Research proposals

PhD selection is performed twice a year through an open process. Further information are available at http://dottorato.polito.it/en/admission.

Grants from Politecnico di Torino and other bodies are available.

In the following, we report the list of topics for new PhD students in Computer and Control Engineering. This list may be further updated if new positions will be made available.

If you are interested in one of the topics, please contact the related Proposer.


- Blockchain for Physical Internet and Sustainable Logistics
- Stochastic optimization problems for Urban Mobility and City Logistics
- Advancing Mobile UI Testing Techniques
- Augmented Reality for Visual Interactive Information
- A wearable platform for exhaustive monitoring of Parkinson’s disease during Ac...
- Key management techniques in Wireless Sensor Networks
- Machine Learning Acceleration in Resource-Constrained Devices for Smart Manufact...
- Big Crisis Data Analytics
- Explainability and Improvisation in the Making of Self-aware Conversational Agen...
- ICT for Urban Sustainability
- Virtual and Mixed reality for medical education and training
- Biometrics in the wild
- Programming Massively Parallel Server and Embedded Processors
- Concept Maps
- Approximate Computing for Power and Performance optimization
- Methods and Algorithms for Efficient Computational Physics-of-Matter of Electron...
- Cross-lingual text mining
- Promoting Diversity in Evolutionary Algorithms
- Visual and Human-Centered Computing for Industry 4.0
- Multivariate analysis and augmented reality visualization in research and indust...
- Human-level semantic interpretation of images and video
- Intuitive machine learning for Smart Data applications
- Development of Vulnerability-Tolerant Architectures
- Improving the dependability and resilience to cyberattacks of next generation co...
- Blockchain for Industry 4.0
- Effects of Data Quality on Software Application Bias and Mitigation Strategies, ...
- Efficient Functional Model-Driven Networking
- Machine learning for sentiment analysis
- Automotive SoC Reliability and Testing
- Accessibility in the Internet of Things
- Recursive set-membership estimation algorithms with application to system identi...
- From data mining to data "storyboarding"
- Urban intelligence
- Visual domain generalization through agnostic representations
- Investigation of innovative IoT technologies for complex manufacturing process m...
- Domain Generalization via Self-Supervised Learning
- First person action recognition from multi-modal data


Detailed descriptions

Title: Blockchain for Physical Internet and Sustainable Logistics
Proposer: Guido Perboli
Group website: http://icelab.polito.it/
Summary of the proposal: The blockchain is an incorruptible digital ledger of economic transactions that can be programmed to record not just financial transactions but virtually everything of value.
Therefore, this Ph.D. thesis wants to explore the potential applications of blockchain to the supply chain problem, in particular when it is associated with the Physical Internet paradigm, a vision for a sustainable and deployable solution to global problems associated with the way “we move, store, realize, supply and use physical objects all around the world”.
In more details, the candidate will explore the usage of the IOTA blockchain, a blockchain developed explicitly for Internet of Things applications.
These activities will be carried out in collaboration with the ICE for City Logistics and Enterprises Lab of Politecnico di Torino and Istituto Superiore Mario Boella ICELAB@Polito). The Ph.D. candidate will be co-supervised by Ing. Edoardo Calia (ISMB). ICELab@Polito is already member of the IOTA Foundation and partner in the creation of the first marketplace for IoT Data with the Municipality of Turin. The student will be integrated in the Blockchain group of the ICELab@Polito. Presently, four master students are already working both to ICT and Business aspects of the Blockchain.
Rsearch objectives and methods: The objectives of this research project are grouped into three macro-objectives:

Management Science/Operations Research objectives:
- Integration of Physical Internet paradigm and Blockchain
- Definition of operations management problems arising from the integration of freight transportation and blockchain

ICT:
- Implementation differences with Bitcoin;
- Deep understanding of ICT infrastructure;
- A better understanding of data privacy, integrity, trustability, sharing, and standardization.

Deployment and field testing:
- Smart City applications;
- Validation of the solutions (in collaboration with ICE Lab).
Outline of work plan: PHASE I (I and II semesters)
- State of the art on the blockchain, Physical Internet, supply chain and transportation in a smart city.
- Identification of needs and integration with City Logistics and Physical Internet.

The student will start his/her research from the state-of-the-art analysis carried by prof. Perboli and ICELab@Polito, presented to the next 9th IFIP International Conference on New Technologies, Mobility and Security in the tutorial of prof. Perboli titled Applications of Blockchain to Supply Chain and Logistics: emerging trends and new challenges

PHASE II (II and III semesters).
- Identification of promising applications;
- Implementation of the blockchain-based applications.

The student will start from the analysis of the current industrial projects and the analysis of the needs coming from the companies and associations collaborating with ICELAb@Polito (mainly, TNT, DHL, Amazon and SOSLog).

PHASE III (IV semester)
- Deployment of solutions in collaboration with ICE Lab;
- On-field test and refinement.

PHASE IV (V and VI semesters)
- Cost-benefit analysis;
- Business model and market analysis;
- Guidelines for standardization of blockchain-based transportation applications;
- Scalability and replicability of the solution on a large scale.
Expected target publications: - Transportation Research;
- Interfaces - INFORMS;
- Omega - The Int. Journal of Management Science - Elsevier;
- Management Science - INFORMS;
- Business Process Management Journal;
- IT Professional - IEEE;
- IEEE Access
- Sustainable Cities and Societies – Elsevier;
- Transportation Research (Part A-E) – Elsevier;
- Transportation Science – INFORMS.
Current funded projects of the proposer related to the proposal: There are currently two proposal H2020 under submission.
Possibly involved industries/companies:Amazon, Deloitte, IBM

Title: Stochastic optimization problems for Urban Mobility and City Logistics
Proposer: Guido Perboli
Group website: http://icelab.polito.it/
Summary of the proposal: The proposal considers the problem of optimizing the services in different application settings, including road network and logistics operations. When dealing with real urban logistics applications, the aforementioned problems becomes large and normally affected by uncertainty in some parameters (e.g., transportation costs, congestion, strong competition, etc.).
Under this context, one of the most suitable approaches is modeling the problem as a stochastic optimization problem.
Unfortunately, solving large-sized bilevel optimization problems with integral variables or affected by uncertainty is still a challenge in the literature. This Research Proposal aims to fulfill this gap, deriving new exact and heuristic methods for integral and stochastic programs.

These activities will be carried out in collaboration with the ICT for City Logistics and Enterprises lab of Politecnico di Torino and the CIRRELT of Montreal. Actually, the PhD candidate will be co-directed by prof. Teodor Gabriel Crainic, who is a worldwide specialist in Urban Logistics.
Rsearch objectives and methods: The objectives of this research project are grouped into three macro-objectives:

Integer Programs:
- Define price-setting models for last-mile logistics operations
- Develop exact and heuristic methods based on the mathematical structure of the problems for solving the aforementioned problems

Stochastic Programs:
- Define stochastic price-setting models for network optimization
- Develop exact and heuristic methods based on the mathematical structure of the problems and the knowledge of the distribution of probability of the uncertainty parameters

Last Mile and City Logistics Instances:
- Test the new models and methods in real Last Mile, E-commerce and City Logistics projects in collaboration with the ICE Lab.
Outline of work plan: The PhD Candidate will develop his/her research in both the research centers (ICELab@POLITO and INOCS). In more detail, he/she will spend about half of the time in INRIA Lille.

PHASE I (I semester). The first period of the activity will be dedicated to the study of the state of the art of Integer Stochastic Programming, as well as of the Last Mile and City Logistics applications.

PHASE II (II and III semesters). Identification of the main methodological difficulties for the design of efficient solution methods for Integer models studied in this proposal. Identification of the model properties (cuts, specific lower bounds, eventual dependencies of the variable values) letting the solution method to converge with a limited computational time to high-quality solutions. A great challenge will come from the combinatorial nature of the problems under study. In fact, the majority of the literature is focusing on continuous or quasicontinuous problems. The problems in logistics require integer and decisional variables, making the problems much more complex to solve. The candidate will then study both exact and heuristic methods. In particular, the candidate will start from the studies of the PhD proposer on stochastic and deterministic packing problems. In fact, it is already proved that these problems can model several real settings in logistics and supply chain.

PHASE III (IV and V semester). Identification of the main methodological difficulties for the design of efficient solution methods for Stochastic models studied in this proposal. Identification of the main properties letting the definition of solution methods converging with a limited computational time to high-quality solutions.

PHASE IV (V and VI semesters). Prototyping and case studies. In this phase, the results of Phase II and III will be applied to real case studies. Several projects are currently in progress, which require experimentations and the validation of the proposed solutions.
Expected target publications: - Journal of Heuristics - Springer
- Transportation Science - INFORMS
- Transportation Research part A-E - Elsevier
- Interfaces - INFORMS
- Omega - The International Journal of Management Science - Elsevier
- Management Science - INFORMS
- Computer and OR - Elsevier
- European Journal of Operational Research – Elsevier
Current funded projects of the proposer related to the proposal: Urban Mobility and Logistics Systems Interdepartmental Lab
Possibly involved industries/companies:TNT, Amazon

Title: Advancing Mobile UI Testing Techniques
Proposer: Marco Torchiano
Group website: http://softeng.polito.it/
Summary of the proposal: Testing of mobile application user interfaces consists in writing test cases that exercise the UI, and allow performing end-to-end testing of the whole app in an automated way. Tests for mobile applications are particularly challenging since they have to address a wide range of devices and platforms.

Despite the presence of several mature frameworks, UI tests show a limited adoption. This is certainly due to the significant effort in writing the tests cases, though the main barrier is represented by the inherent fragility of such kind of tests: a minor change in the UI can “break” the test requiring a significant effort in identifying the cause and then fixing the test.

Lowering the adoption barrier requires both working on existing test suites with the ability to detect fragile tests in existing suites and then techniques to fix such fragilities; in addition novel approaches to build tests ought be devised that minimize writing effort, and enable effective strategies to evolve test suites.
Rsearch objectives and methods: O1.Identification of test fragilities

This step is about the definition of techniques to detect the patterns that cause test fragility. A comprehensive taxonomy (O1.1) is the prerequisite. The automatic detection of fragilities (O1.2) can then be developed. A tool that can work as a plug-in of an IDE represents the main outcome.

O2.Test fragility mitigation

Building upon the results of O1, we need to identify remediation techniques for the fragility-inducing patterns (O2.1). The remediation techniques must be implemented in order to achieve automatic removal (O2.2). The implementation should find its place within a plug-in for an IDE.

O3.Definition of novel testing techniques

The identification of the fragility-inducing patterns represents the basis also for a gap analysis of existing techniques (O3.1). Novel techniques should be defined (O3.2) typically by leveraging the relative strengths of existing ones.

O4.Definition of test evolution strategies

The key limitation of UI tests is represented by the need for evolving along with the application UI. A set of recommendations and strategies must be devised to guide the evolution of both existing and novel tests.
Outline of work plan: The main activities conducted in the three years of the PhD studies are:
Task 1.1: Construction of a defect taxonomy and relative validation
Task 1.2: Development of a fragility detection tool
Task 1.3: Empirical assessment of effectiveness of fragility detection
Task 2.1: Identification of the fragility mitigation techniques
Task 2.2: Development of a fragility removal tool
Task 2.3: Empirical assessment of fragility removal
Task 3.1: Analysis of limitation of existing tools and techniques
Task 3.2: Definition of a novel (integrated) testing approach
Task 4.1: Identification of a test evolution strategy
Task 4.2: Empirical assessment of test evolution strategy

The four work units will follow similar methodological approaches where
- an initial identification of the issues is conducted by means of observational methods (case studies, surveys, case control studies);
- a solution is proposed
- a proof of concept implementation is developed
- an empirical assessment is conducted typically using controlled experiments.
Expected target publications: The target for the PhD research includes a set of conferences in the general area of software engineering (ICSE, ESEM, EASE, ASE, ICSME) as well as in the specific area of testing (ICST, ISSTA).
More mature results will be published in software engineering journal, the main being: IEEE Transactions on Software Engineering, ACM TOSEM, Empirical Software Engineering, Journal of Systems and Software, Information and Software Technologies.
Current funded projects of the proposer related to the proposal: There is an Italian PRIN proposal under evaluation
Possibly involved industries/companies:Several companies expressed concerns on mobile UI tests, among them three showed more interest in the specific goals of this PhD proposal: Reale Mutua Assicurazioni, Intesa SanPaolo, Engineering

Title: Augmented Reality for Visual Interactive Information
Proposer: Andrea Sanna
Group website: http://grains.polito.it/index.php
Summary of the proposal: This Ph.D. project will explore novel methods and interfaces to display and interact with (possible) highly dynamic data by AR technologies. A new class of systems, applications and interaction techniques designed to support users with visual interactive information on the go will be designed and implemented. These explorations will involve the creation of new visualizations and interaction techniques, the development of prototype systems that encompass emerging technologies, and evaluation through user studies. A special focus will be devoted to design AR application for enhancing the user experience when playing games and serious games.
Rsearch objectives and methods: AR allows computer generated information to be overlaid on the real world. Technologies related to AR are advancing quickly, and recent developments include fully-self-contained, wearable systems (e.g., the Microsoft Hololens and Meta2). As wearable devices progress, they will become invaluable to a variety of users who will benefit from easy access to information. For instance, construction workers may view future building plans superimposed on a current job site and workers roaming an industrial plant can bring a virtual control room wherever they go. Moreover, video game players may live completely new experiences without spatial constraints involved in traditional games. For all these applications, researchers and scientists have to face new challenges related to the visualization and the interaction with highly dynamic information.

This project will investigate potential uses of the AR displays for visualizing and interacting with dynamic data. The project will explore the relationship between visual augmentation and spatial cognitive function, which needs to meet the required level to be able to gain the understanding of the meanings of the targeted data. The project will design and develop AR applications to visualize both scientific data (such as molecular structure, climate changes, and environmental sensor data) and other kind of data related to playful applications such as games (e.g., interactive maps, multiple enemies/targets, and so on). A suit of novel intuitive spatial interaction techniques will be proposed, developed, and evaluated to fit the application needed. The outcomes of this project have to bridge the gap between highly interactive and dynamic data generated by nomadic applications and augmented reality technologies.
Outline of work plan: The work plan will be defined year by year; for the first year of activity, it is expected the candidate will address the following points:
- Analysis of the state-of-the-art (in this phase, both AR technologies (SW and HW) and the existing application to display multivariate date have to be investigated. The candidate could/should complete this analysis step by attending to specific Ph.D. courses related to these topics (HCI, data analytics, information visualization, etc.).
- Objective definition (after the analysis step, the candidate has to be able to identify challenging open problems, thus defining concrete objectives for the next steps).
- Methodology (concurrently to the objective definition step, the research methodology has to be also chosen; a right mix between theoretical formulation and experimental tests has to be identified).
- KPIs definition (the definition of clear, objective and measurable parameters plays a key role for the assessment of this kind of applications/interfaces). - Expected outcomes (research papers, scientific contributions, significance, potential applications).

The next years will be mainly focused on the design, implementation and test phases.
Expected target publications: IEEE Transactions on Human-Machine Systems
IEEE Transactions on Visualization and Computer Graphics
ACM Transaction Human Computer Interaction
International Journal of Human–Computer Interaction
Current funded projects of the proposer related to the proposal: HuManS (bando fabbrica intelligente)
Possibly involved industries/companies:COMAUN, FCA, CRF, CNH

Title: A wearable platform for exhaustive monitoring of Parkinson’s disease during Activities of Daily Living
Proposer: Gabriella Olmo
Group website: http://www.sysbio.polito.it
Summary of the proposal: In Parkinson's disease, patients suffer from disabling motor symptoms such as bradykinesia, resting tremor, rigidity and postural instability. L-dopa controls these symptoms in an early stage of the disease. However, with the progression of the condition, L-dopa becomes less effective, while inducing side effects such as dyskinesia. Motor response fluctuations (end-dose washout, on-off periods, and freezing of gait) greatly impair the patient’s quality of life. Both dyskinesia and motor fluctuations may be coped with by fine-tuning the drug posology on every single patient. However, symptoms are difficult to appreciate in a medical office, and their assessment is currently limited to poorly reliable self-reports. Moreover, presently the outpatient schedule is once per year, which is suboptimal. Only long-term monitoring, possibly carried on during activities of daily living, can provide a reliable assessment of motor fluctuations.

In this proposal, we want to design a tool for identification and continuous tracking of motor fluctuations in PD patients, using low-power, low-cost, wearable inertial sensors, in line with the modern paradigm of personalized medicine. We intend to use this platform to support neurologists in clinical trials (e.g. administration of opicapone) and in assessing candidacy to surgical therapies (Deep Brain Stimulation), as well as to help in managing DBS/Duodopa devices.
Rsearch objectives and methods: Parkinson's disease (PD) is the second most common neurodegenerative disorder, with a prevalence exceeding 1.9% over the age of 80. Patients exhibit both motor and non-motor symptoms related to the degeneration of dopamine neurons in the brainstem area called substantia nigra. The cardinal motor symptoms are bradykinesia, resting tremor, rigidity and postural instability. Non-motor symptoms include olfactory impairment, orthostatic hypotension, constipation, sleep disturbances, speech problems.

Levodopa (L-dopa), a dopamine precursor that can cross the blood-brain barrier, remains the gold standard therapy for controlling PD motor symptoms. Yet, in advanced stages, L-dopa modifies the natural evolution of PD inducing involuntary movements (dyskinesia) as well as motor response fluctuations (end-dose washout, on-off periods, freezing of gait, nocturnal akinesia). For example, freezing of gait (FOG) is a form of akinesia defined as a brief, episodic absence or marked reduction of forward progression of the feet despite having intention to walk. It occurs during the initiation of gait, and may be elicited by activities such as turning, passing though narrow spaces, negotiating obstacles or performing dual tasks. FOG provokes a high risk of falls, reduces functional independence and impairs quality of life. FOG correlates with motor fluctuations and progression of the disease; the knowledge about its frequency, duration, daily distribution and response to drug therapy is crucial for a reliable patient's assessment

Motor fluctuations can be faced adjusting the drug posology to match each patient's response. However, they can seldom be appreciated in a medical office, and their assessment is currently limited to poorly reliable self-reports. For example, the episodic nature of FOG makes it difficult to catch events during the brief, pre-scheduled follow-up sessions. The FOG occurrence is related to the time elapsed since the last L-dopa administration, is dependent on the patient's attention devoted to gait, and on many subtle cognitive factors. It turns clear that only long-term observation, possibly carried on during activities of daily living (ADL), can provide a reliable assessment of this phenomenon. Similar considerations hold for aspects such as daily evolution of bradykinesia and postural instability.

In this research, we want to design a tool enabling the identification and continuous tracking of motor fluctuations in PD patients, using low power and low cost wearable inertial sensors. By means of acceleration data, we plan to recognize the patient’s activities, and to get information about his/her motor status. The ultimate goal is to provide the clinicians with a complete clinical picture of each patient. This is in line with the modern paradigm of personalized medicine, which makes use of the several data related to a patient and employs proper classification methods to provide effective and supportive treatment, accessible to all patients in a context of overall cost reduction. We intend to use this device to support neurologists in clinical trials (e.g. administration of opicapone, a new drug designed to curtail motor fluctuations) and to evaluate motor fluctuations and dyskinesia for assessing candidacy to surgical therapies (Deep Brain Stimulation or Duodopa dispenser). The main challenge will be on finding suitable algorithms for analyzing sensor data, achieving an optimal trade-off between precision, cost, and battery usage.
Outline of work plan: Goal: to develop a tool for PD motor fluctuation monitoring, using inertial data acquired from a wearable device, during ADL in unsupervised manner. The tool is conceived as a sort of electronic diary replacing the patient's self-reports. A non-exhaustive list of items to be monitored is: FOG: number, duration and daily distribution of episodes. As a preliminary assessment, we have already collected more than 3.5 hours of acceleration data on 90 patients, wearing a waist-mounted smartphone.
Bradykinesia. PD is clinically assessed by means of scores such as MDS-UPDRS (Unified Parkinson's Disease Rating Scale). We want to emulate proper items of UPDRS related to bradykinesia. We will focus on Leg Agility (LA), recognized to be strictly correlated to bradykinesia, but we plan to extend the research to other items.
Postural Instability: Also included in MDS-UPDRS, postural instability is a key point in PD patient management. We want to classify the level of risk of each patient, its evolution in time and its correlation with drug posology.

For each item, we plan to design detection and classification algorithms with proper sensitivity and specificity. The software will be ported on STMicroelectronics' SensorTile, a tiny IoT module leveraging a STM32L476JGY microcontroller, Bluetooth low energy connectivity and a wide spectrum of motion and environmental MEMS sensors.
Workpackages:
- Implementation of proper ML/deep learning algorithms (Ist year).
- Performance evaluation on patients recruited at the Molinette Hospital in semi-supervised conditions (Ist year).
- Selection of the best algorithms and testing at patient’s home during ADL (IInd year).
- Porting on the SensorTile platform (IInd year)
- Testing and refinement (IInd year)
- Use in clinical trials to be agreed with neurologists at the Molinette hospital (e.g. opicapone or DBS tuning experiments). (IIIrd year)
Expected result: PD monitoring system in ADL, using wearable sensors. Protocol for data exchange with neurologists. Use of the device in clinical trials. Possible use of the device to tune Duodopa dispenser and/or DBS frequency of stimulation.
Expected target publications: We plan to have at least two journal papers published per year.
Target journals:
IEEE Transactions on Biomedical Engineering
IEEE Journal on Biomedical and Health Informatics IEEE Access
IEEE Journal of Translational Engineering in Health and Medicine
Elsevier Journal "Parkinsonism and Related Disorders"
Elsevier Journal "Gait and Posture"
Current funded projects of the proposer related to the proposal: Not yet
Possibly involved industries/companies:At present, the Regional Expert Center for Parkinson's disease and Movement Disorders of the Molinette Hospital, coordinated by prof. Lo Piano, cooperates with our Department in a feasibility study involving more than 100 PD patients. The Center has expressed interest in the preliminary results and the will of continuing the cooperation.

STMicroelectronics has expressed the intention of making available for the project their proprietary platform SensorTile www.st.com/en/evaluation-tools/stevastlkt01v1.html?ecmp=tt7201_us_social_may2018

Title: Key management techniques in Wireless Sensor Networks
Proposer: Filippo Gandino
Group website: http://www.cad.polito.it/
Summary of the proposal: Wireless sensor networks (WSNs) offer benefits in several applications but are vulnerable to various security threats, such as eavesdropping and hardware tampering. In order to reach secure communications among nodes, many approaches employ symmetric encryption. Several key management schemes have been proposed in order to establish symmetric keys exploiting different techniques (e.g., random distribution of secret material and transitory master key). According to the different applications of WSNs, the state-of-the-art protocols have different characteristics and different gaps.

The proposed research activity will be focused on the study and development of key management protocols. The proposed investigation will consider both the security requirements and the performance (e.g., power consumption, quantity of additional messages, computational effort) of the networks. The research will analyze different possible solutions, evaluating the trade-off in terms of costs and benefits, according to different possible scenarios and applications.

The initial part of the activity will be focused on the proposals currently in progress within the research group. However, the candidate will be stimulated to formulate and develop new solutions.

The proposed research involves multidisciplinary knowledge and skills (e.g., computer network and security, advanced programming).
Rsearch objectives and methods: The objectives of the proposed activity consist in studying the state-of-the-art of key management protocols for WSNs and to propose new solutions suitable to various application scenarios. The requirements that affect the security protocols will be considered in order to find the best solution for different kinds of WSNs. In particular, the mobility of the nodes and the possibility to add new nodes after the first deployment will be considered.

According to the current trends in this research field, the starting activity will be focused on the use of combinatorial approaches for the distribution of the keys. The use of these techniques is considered appealing, since the provided security properties are more effective than the ones based on randomness. However, they involve drawbacks in terms of memory overheads and limits to the size of the network.

Moreover, the research activity will consider key management in network dedicated to specific applications. In particular, the first considered case study will be the Internet of Things in which WSNs represent an important building block and require specific solutions for an effective interoperability.

The candidate will start to cooperate to the ongoing research activity by implementing, testing and evolving the proposals actually under consideration by the research group. However, a fundamental goal of the candidate is to reach skills suitable for the development of autonomous proposals.

The proposed solutions will be evaluated and compared to state-of-the-art approaches, in order to evaluate their security level and their performance. The analysis of the protocols will be composed by: (a) a theoretical analysis, (b) simulations, and (c) implementation on real nodes.

The theoretical analysis will consider several aspects of the key management protocols. A part of this analysis will evaluate the statistical characteristics of the scheme, in order to analyze the level of security and other network parameters for protocols based on stochastic elements. A security analysis will compare the properties and the requirements of the protocols.

The simulations will be used to analyze the performance of the schemes with a high number of nodes. This kind of investigation allows reaching significant results within a limited time. There are many available tools that allow simulating WSNs, e.g., TOSSIM, which uses the same code written for the TinyOS platform.

The last kind of analysis will be based on an experimental session on a real network. The main platform that will be used will be TinyOS, in order to develop a code that can be used also for the simulations. The main purpose of a real implementation is to validate the results achieved by simulations.

In order to work on this research topic, a candidate may have good programming skills, and he/she should have a good knowledge on networking and security.
Outline of work plan: During the first part of the first year the candidate will investigate the field of the research activity. Therefore, the candidate will study the existing protocols and will improve his/her knowledge on the related topics. This phase lasts 3 months. During the rest of the first year, the candidate will start to work on research proposals currently under consideration within the research group. The PhD student will cooperate in the detailed definition of the key management schemes, and in their implementation, considering a real Internet of Things case study WSN. The preliminary results will be submitted to international conferences.

During the second year, an accurate analysis and evaluation of the proposed ideas will be done, and an eventual modification of those approaches will be considered. This task will include a programming activity, in order to test the proposed solutions and to integrate them in real applications. The achieved results will be submitted to an international journal.

During the last year, the main goal of the candidate will be to conclude the analysis on his proposals and to publish the results. However, also the task of evolution of the current proposals of the research group will be curried on, in order to differentiate its research activity and to increase the probability of reaching valuable results.

During all the three years the candidate will have the opportunity to cooperate in the development and implementation of key management solutions applied to research projects. These tasks will allow the candidate to understand the real scenarios in which WSNs are used and to find their actual issues.
Expected target publications: IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS
WIRELESS NETWORKS (Springer)
Current funded projects of the proposer related to the proposal: FDM - Bando Fabbrica Intelligente
Possibly involved industries/companies:No

Title: Machine Learning Acceleration in Resource-Constrained Devices for Smart Manufacturing Applications
Proposer: Massimo Poncino
Group website: http://eda.polito.it
Summary of the proposal: An often underrated issue in machine learning (ML) frameworks is their computational complexity: typical ML tasks may require GBs of memory and billions of floating point operations (GFLOPS). The consequence is that these tasks are normally run either on the cloud or off-cloud using complex hardware accelerators such as GPUs.

'Distance' between the source of the data (i.e., sensors in the case of industrial applications) and the computational hardware is critical, both because it introduces latencies that prevent real-time classification, and because it implies the transmission of a large amount of raw data (typically wireless), which results in high energy consumption. Therefore, while offloading part of the computations to the cloud or to custom hardware is a known solution that can help catching up the computational gap, it still does not fully solve the latency and energy efficiency issues. To this end, classification should be carried on local devices (the edge nodes).

The objective of this research is study and implement ML (possibly based on deep learning) techniques that are suitable for implementation of resource-constrained devices such as small embedded boards or mobile devices. An important part of the research will be devoted to specializing the techniques to the specific application domain, i.e, understanding which ML strategy offers the best accuracy/complexity tradeoff for a given classification task.
Rsearch objectives and methods: The final objective of the project is the development of a framework for implementing machine learning tasks on the resource-constrained hardware of IoT edge nodes. The main application domain to which this framework will be applied is the monitoring of processes and equipment in a smart manufacturing environment.

To this end, quality-scalable design techniques will be employed. These techniques aim at realizing software or hardware in which the amount of computational “effort” can be traded off with the quality of the final outputs, possibly at runtime in response to changes in the input data or context. In the specific case of ML tasks, the quality of outputs is represented by the classification or regression accuracy, while a reduction of the computational effort translates into smaller latency and energy consumption for the edge nodes.

Quality-scalable design techniques applicable to the domain in question involve hardware, software and algorithm design.

At the hardware and software level, approximate computing strategies such as precision scaling or loop perforation can be leveraged to (dynamically) reduce the complexity of an inference task, e.g. the forward propagation step of a deep neural network (DNN) both in terms of number of operations and of memory footprint.

At the algorithm level, quality scalability can be achieved by tailoring the classification/regression model to the target problem, context and input data. To this end, the framework will include design exploration techniques to identify the best model(s) in terms of accuracy versus complexity tradeoff for the target application. Based on the results of this exploration, the final deployed model may also be generated as a synergistic combination of ML algorithms of different complexity and accuracy (e.g. a simpler SVM versus a low-cost DNN, or even two DNNs with a different setting of hyperparameters). The most appropriate algorithms will be then automatically selected at runtime based on the characteristics of the inputs, or on the current usage scenario (e.g. low versus high probability of failure).

While some of these techniques have been already applied individually (mostly to image classification problems), there is not yet a complete framework for implementing ML tasks on resource-constrained hardware, leveraging quality-scalable design. Moreover, the smart manufacturing domain poses novel and interesting issues. For example, the optimal ML models to use for the required classification tasks may differ significantly from those involved in image-related problems (e.g. convolutional neural networks).

The candidate will mostly work in C/C++ and Python. A basic knowledge of these languages will help, but is not strictly required. Although this is not foreseen as the main subject of the project, the candidate may also experiment with or modify some digital hardware designs written for ASIC or FPGA targets.
Outline of work plan: PHASE I (months 1-9):
- State of the art analysis
- Characterization of typical sensor data
- Characterization of the categories of inference problems addressed in Smart Factory applications

PHASE II (months 9-18):
- Identification of application-specific simplifications for novel approximate inference architectures
- Implementation of existing inference networks on smartphone-like devices and smaller embedded boards

PHASE III (months 18-36):
- Implementation of the novel approximate inference techniques
- On-line validation of the methods with possible applications in an real-life industrial environment
Expected target publications: E.g.:
- IEEE Transactions on Computers
- IEEE Transactions on CAD
- IEEE Design and Test of Computers
- ACM Transactions on Embedded Computing Systems
- IEEE Transactions of Design Automation of Electronic Systems
- IEEE Pervasive Computing
- IEEE Journal on Internet of Things
Current funded projects of the proposer related to the proposal: - Regional Project "DISLOMAN" (Dynamic Integrated ShopfLoor Operation MANagement for Industry 4.0)
- Project "Definizione e sviluppo di una piattaforma software per la progettazione ed il supporto della produzione"
Possibly involved industries/companies:Ph.D. Scholarship funded by Comitato ICT (Reply)

Title: Big Crisis Data Analytics
Proposer: Paolo Garza
Group website: http://dbdmg.polito.it/
Summary of the proposal: Society as a whole is increasingly exposed to natural disasters because extreme weather events, exacerbated by climate change, are becoming more frequent and longer. To address this global issue, advanced data analytics solutions able to cope with heterogeneous big data sources are needed. The big amount of data generated by people and automatic systems during natural hazard events (e.g., social network data, satellite images of the affected areas, images generated by drones), is typically referred to as Big Crisis Data. To transform this overload of heterogeneous data into valuable knowledge, we need to (i) integrate them, (ii) select relevant data based on the target analysis, and (iii) extract knowledge in near-real time or offline by means of novel data analytics solutions. Currently, the analysis is focused on one single type of data at a time (e.g., social media data or satellite images). Their integration into big data analytics systems capable of building accurate predictive and descriptive models will provide effective support for emergency management.

The PhD candidate will design, implement and evaluate big data analytics solutions able to extract insights from big crisis data.

An ongoing European research project will allow the candidate to work in a stimulating international environment.
Rsearch objectives and methods: The main objective of the research activity will be the design of data mining and big data analytics algorithms and systems for the analysis of heterogeneous big crisis data (e.g., social media data generate by citizens and first responders during natural hazards, satellite images or drone-based images collected from area affected by emergency events), aiming at generating predictive and descriptive models.

The main issues that would be addressed are the followings.

Scalability. The amounts of big crisis data are significantly increased in the last years and some of them are singularly large (e.g., the satellite images). Hence, big data solutions must be exploited to analyze them, in particular when historical data analyses are performed.

Near-real time constraint. To effectively tackle natural hazards and extreme weather events, timely responses are needed to plan emergency activities. Since large amounts of streaming data are generated (e.g., social data, environmental measurements) and their integration and analysis is extremely useful for planning emergency management activities, scalable streaming systems able to process in near-real time data sources and build incrementally prediction data mining/machine learning models are needed. The current big data streaming systems (e.g., Spark and Strom) provide limited support for incremental data mining and machine learning algorithms. Hence, novel algorithms must be designed and implemented.

Heterogeneity. Several heterogonous sources are available. Each source represents a different facet of the analyzed event and provides an important insight about it. The efficient integration of the available spatial data sources is an important issue that must be addressed in order to build more accurate predictive and descriptive models.
Outline of work plan: The work plan for the three years is organized as follows.

1st year. Analysis of the state-of-the-art algorithms and data analytics frameworks for big crisis data storage and analysis. Based on the analysis of the state-of-the-art, pros and cons of the current solutions will be identified and preliminary algorithms will be designed to optimize and improve the available approaches. During the first year, descriptive algorithms, based on offline historical data analyses, will be initially designed and validated on real data related to one specific hazard (e.g., floods) to understand how to extract fruitful pattern for medium- and long-term hazard management/planning actions.

2nd year. The design of incremental and real-time predictive models will be addressed during the second year. These systems will allow managing emergency in near-real time. For instance, classification algorithms will be designed to automatically classify tweets in informative/non-informative in real-time during hazards in order to send to the first responders only useful knowledge.

3rd year. The algorithms designed during the first two years will be improved and generalized in order to be effectively applied in different hazard domains (e.g., floods, fires, heat waves).

During the second/third year, the candidate will have the opportunity to spend a period of time abroad in a leading research center.
Expected target publications: Any of the following journals:
- IEEE Transactions on Big Data (TBD)
- IEEE Transactions on Knowledge and Data Engineering (TKDE)
- IEEE Transactions on Emerging Topics in Computing (TETC)
- ACM Transactions on Knowledge Discovery in Data (TKDD)
- Journal of Big Data
- Big Data Research

IEEE/ACM International Conferences
Current funded projects of the proposer related to the proposal: I-REACT - "Improving Resilience to Emergencies through Advanced Cyber Technologies", H2020 European project (http://www.i-react.eu/)
Possibly involved industries/companies:

Title: Explainability and Improvisation in the Making of Self-aware Conversational Agents
Proposer: Maurizio Morisio
Group website: http://softeng.polito.it
Summary of the proposal: Traditional user interfaces, although graphically rich, are based on a rigid set of commands, with intelligence and flexibility on the user side only. Recently, conversational assistants have been introduced to understand commands in natural language and provide answers. Numerous research initiatives investigated the personalization of the interactions with the assistants paving the way to a closer integration between natural language processing methodologies and recommender systems. Despite these efforts that lifted the assistant to the grade of an advisor, the intelligence is still far from the concept of a real conversational agent. The state of the art is quite initial, with assistants capable of correctly understanding the concepts and intent in free text questions with very low accuracy and precision, in very focused and standardized domains. Even worse is their performance on producing relevant answers, and low robustness across different languages. The automated understanding of the user requests and the retrieval of the related information in order to provide a human-like answer are as of today two outstanding challenges. We propose a research investigation to improve the capabilities of assistants up to real conversational agents, working both on natural language understanding and answer generation on English and Italian text focusing on two human-based intelligence assets: explainability and improvisation.
Rsearch objectives and methods: The research is centered around the two main problems, understanding requests/questions from the user in natural language (Objective 1), then computing relevant and aware answers (Objective 2). In parallel a demonstrator is built to evaluate the approach (Objective 3).

Objective 1: Understanding. Design of a deep learning approach for the automated and context-tailored understanding of user requests across different contextual domains. State-of-the-art techniques propose approaches based on recurrent neural networks trained with word embeddings and domain-specific training data. We will build upon these approaches extending them with the concept of thread, i.e. sequences of requests and answers, pivoting the time component and sentence dependence. We aim to reach better performance computed as F-measure over broader domains (beyond DBpedia), and on both English and Italian. We remark that working with the Italian language is harder because of the more limited availability of open source tools and data.

Objective 2: Generation. Design of a knowledge-based approach for the retrieval and filtering of information exploiting domain-specific knowledge graph embeddings. State-of-the-art approaches propose machine learning popularity based (PageRank) mechanism for the retrieval of information from databases and the generation of a template-based answer. We will build upon these and introduce the notion of salience within a given context that is represented in knowledge graphs. We aim to reach better performance computed as F-measure over broader domains (beyond DBpedia), and on both English and Italian and make the agent being able to explain what it has generated and aware of the notion of improvisation.

Objective 3: Runnable prototype (TRL 3) that is able to hold a conversation with both a user providing tailored domain-specific requests in English and Italian.

In terms of research results vs the state of the art we aim at
- better precision and recall in entity and intent recognition (objective 1), in answer generation (objective 2)
- over wider knowledge domains, thus generating knowledge graphs domain-specific
- for both English and in Italian
- with a set of observable metrics to define explainability and improvisation
Outline of work plan: Year 1 (objective1, objective3): Investigation and experimentation of deep learning techniques exploiting knowledge graphs for intent and entity classification from natural language text.

Year 2 (objective1, objective3): Investigation and experimentation of deep learning techniques exploiting knowledge graphs for content-based dialogue understanding in order to personalize the processing of the requests according to user profiles and contexts, generating non-obvious answers, which can be explained. Will exploit domain-specific knowledge graphs (such as tourism, scientific literature, encyclopedia) and process both English and Italian.

Year 3 (objective2, objective3): Investigation and experimentation of information retrieval mechanisms to intelligently tapping into domain-specific knowledge graphs (such as tourism, scientific literature, encyclopedia) and to generate non-obvious natural language answers over unseen topics.

During the 3 years, we will run continuously in-lab benchmark validation of the approach using well-known gold standard compared with state-of-the-art approaches.

The candidate will be also actively involved in an international project and a national project. Will be co-tutored by Politecnico di Torino and LINKS Foundation, working in a unique environment created by the blend of an academia and a research center. In addition, there will be the possibility that candidate will spend 1 year abroad in a top-tier research center.
Expected target publications: 3 top-tier conference papers such as WWW, ISWC, RecSys, ESWC, ACL, LREC
1 journal such as Semantic Web Journal, Knowledge-Based Systems, Intelligent Systems, Information Processing and Management
Current funded projects of the proposer related to the proposal: PasTime (EIT Digital no. 1716)
Possibly involved industries/companies:Amadeus, TIM, Links

Title: ICT for Urban Sustainability
Proposer: Maurizio Rebaudengo
Group website: http://www.cad.polito.it/
Summary of the proposal: Air quality nowadays is receiving ever growing attention since it has become a critical issue, in fact long-term exposure to polluted air can result in permanent health issues. The causes of air pollution are various: fossil fuels, power plants, factories, waste incineration, controlled burn, cattle farming and so on. All these sources contribute to release several toxic agents. In fact, they can put in danger not only the people - it is one of the major cause of deaths per year - but also our entire ecosystem. The research will investigate a smart city system to help local government and citizens themselves to monitor what currently happens in the city. The designed architecture of the system will use sensor networks able to capture city condition like temperature, air pollution, traffic situation, etc. The system will be deployed in a mobile and static network installed on different entities, like the bicycles available in the bike-sharing service, the public transportation vehicles and fixed spots to acquire real-time and diffused information on the urban area. The huge amount of data will be analyzed to develop mobile applications useful to help the citizens and the stakeholders in the management of the actions to reduce the air pollution or the traffic congestion, and optimizing the use of the urban resources. The research activity will be focused on some open issues related to the efficient application of mobile wireless sensor networks: e.g., the evaluation of the mobile systems in case of deterministic and stochastic routes, the auto-calibration of data collected by low cost sensors, the energy saving based on optimization of transmission and data collection protocols, the application of machine learning approaches to selectively discard erroneous or not relevant data.
Rsearch objectives and methods: The research will study and develop a hardware system and a software tool able to monitor, control and handle urban problems such as air pollution, traffic congestion, water quality, and so on.

The research will design the whole architecture of the system composed of different types of nodes, fixed and mobile. Mobile nodes could be deployed into public transport vehicles and public bicycles. Fixed nodes will be installed in some critical points, like the most crowded crossroads. The use of mobile nodes requires a study on the spatial and temporal coverage due to the possible route strategies (e.g., stochastic routes for bike sharing, fixed paths for busses), in order to verify the trade-off between costs and benefits for an efficient urban air quality monitoring.

Among the amount of hazardous gases, some pollutants (e.g., CO, ground level O3, Particulate Matter, and Pb) are the most dangerous and will be considered as the starting point for the development of the sensor nodes. The use of low cost sensors involves a temporal variation of the sensor output. In order to address this problem, algorithms for the auto-calibration of the data will be studied and developed. In case of mobile node, the localization of each node becomes an essential constraint, in order to realize an ubiquitous real-time monitoring system: the integration of a GPS module, which delivers accurate position and time information, will be considered in order to guarantee a pervasive system. A particular attention will be paid in evaluating the scalability of the system, in order to maximize the number of possible nodes. Possible integration with different technologies will be considered, and in particular RFID-based sensors fixed nodes will be exploited in cooperation with mobile WSN nodes. Data will be exchanged among the nodes and toward a gateway exploiting low-power wireless data routing protocols or a telecommunication network. The research activity will consider the optimization of data acquisition, storing, aggregation and transmission, in order to reduce energy consumption. Data will be processed on the low-level module of the network and then will be collected by distributed gateways and a central server according to a hierarchical architecture that must guarantee their reliability and the availability of the system.

The foreseen activity will consider also the analysis of the huge amount of data with the goal to realize a platform of data acquisition and evaluation, useful to develop mobile applications to be installed in portable or automotive devices, with the goal to display a dashboard containing useful information such as air pollution, traffic condition or parking availability, and to elaborate those information to make some decision useful for the user citizen, e.g., the best path, as far as the traffic is considered, the area with a high or low level of pollution, the zone with available parking, etc.
Outline of work plan: The research activity is organized in 3 main work packages (WPs). The first WP includes the design, implementation and experimentation of heterogeneous devices for monitoring useful parameters. The second WP deals with their integration in a large WSN for a widespread coverage of the city. Finally, the third WP consists in the management and analysis of the data collected from the WSN.

The first activity of WP1 aims to identify, for each urban problem that will be faced, a set of parameters for a complete characterization. For example, traffic congestion can be evaluated recording the number of passages in fixed points, air pollution can be measured according to the level of noxious gases in the air. It is expected that different modules will be integrated to obtain more comprehensive information: for example, a GPS module can be added to a sensor for precisely localize the place of measurement. State-of-the-art protocols for data acquisition, storing, aggregation and transmission will be studied and implemented on the nodes in order to find new optimized energy –saving solutions. In the last part of WP1, the developed devices will be tested on the laboratory and on the field to evaluate their effectiveness in facing the considered urban problems. A special attention will be focused on the auto-calibration of the data, aiming at proposing new solutions in order to improve the accuracy of the collected data.

The activities of WP2 are based on an initial state-of-the-art analysis to identify the most common practical issues in the design of a big WSN, with a special focus on their adoption in smart cities. Existing solutions will be reviewed in order to propose an effective integration of the devices in a robust network. The proposed infrastructure will be implemented and tested. Particular attention will be paid to the efficiency and security of the WSN, in terms of resource consumptions and communication reliability and security. A study on the characteristics of possible vehicles for the implementation of a mobile sensors network will investigate the time and spatial frequency of the measurements, the additional cost (e.g., in terms of human labor) for the system and the possible alteration of the data (e.g., due to the emissions of the vehicle).

The first activity carried on during WP3 is the development of a database for storing the data acquired from the WSN and for efficiently retrieving this information: Data Mining algorithms will be studied and developed for a deep analysis of the measured data; the main activity of WP3 regards the development of dashboard useful to develop solutions exploiting the so-formed knowledge for effectively improving the urban life.

All the results achieved in the WPs will be submitted to international conferences and journals.
Expected target publications: IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
IEEE INTERNET OF THINGS JOURNAL
WIRELESS NETWORKS (Springer)
JOURNAL OF NETWORK AND COMPUTER APPLICATIONS (Elsevier)
Current funded projects of the proposer related to the proposal: Progetto FDM, Regione Piemonte
Possibly involved industries/companies:Città di Torino, Fondazione Torino Smart City, 5T, ARPA

Title: Virtual and Mixed reality for medical education and training
Proposer: Andrea Bottino
Group website: http://www.polito.it/cgvg
Summary of the proposal: Virtual and Mixed Reality (VMR) are transforming the way medical students and healthcare providers are learning. The training of medical/surgical procedures requires high levels of perceptual, cognitive and sensorimotor skills. Medical education encompasses basic knowledge acquisition, operation of equipment, communication skills with patients and much more. The standard approach to master such complex skills relies on workplace or class learning. However, these practices can involve safety, cost and didactic issues. Here is where VMR technologies can provide a valuable contribution. VR allows creating simulated 3D interactive scenario, providing entertaining (and safe) experiential learning environment where individuals can practice skills and procedures. MR allows augmenting user perception with valuable information about the educational context. VMR simulations facilitate self-directed learning, as trainees can develop skills at their own pace with unlimited repetition of specific scenarios, do not require the presence of an instructor (thus, helping reduce costs) and support teaching institutions and learners with standardized computer-based assessments. However, fully exploiting the possibilities offered by VMR in medical education requires addressing a number of important research questions related to usability, learning paradigms, management of team activities and validation of the developed approaches.
Rsearch objectives and methods: The objective of this research proposal is to tackle the following challenges, proposing and validating novel and promising solutions that can positively affect the quality and the effectiveness of learning.

Usability and User eXperience (UX). When exploiting VMR for training and education in a medical context, the target end users are heterogeneous and likely to have few (or null) previous experiences with the specific technologies. This poses a challenge for the UI designer, since potential users must be enabled to operate the system efficiently in the shortest time possible. Thus, the UI design should reduce the user’s cognitive load and foster learnability, intuitiveness and ease of use. It is worth noting that improving usability and UX can also help soften possible problems due to resistance to tech from end users.

Team activities. In medical context, many of the learning and training activities must be performed in teams, whose members jointly cooperate to the management and execution of the procedures. To this end, the spatial dimension and multimodal capabilities of multi-user VMR environments offer new possibilities for designing, deploying and enacting collaborative activities. However, fully exploiting these potentials requires facing several challenges.
First, enabling an effective user collaboration might require the design and deployment of shared environments capable of:
(1) guaranteeing multiple forms of collaboration among users (i.e., co-located, distributed or mixed),
(2) enabling a seamless integration of multiple interaction and visualization devices (in both MR and VR) and,
(3) supporting the adaptation of the visualized information to the different user roles (e.g., doctors, caregivers and patients should be provided with different views of the same object, or with different information associated to it, and should be allowed to interact with the same object in different ways).
Second, in shared environments, the available interfaces affect the way users collaborate with each other and organize their joint activities in the common (co-located or remote) working space. This raises a variety of empirical questions and technical issues: how do users coordinate their actions? How can they monitor the current state and activities of other users? How to establish joint attention? Which interaction paradigms are most effective for groups of local or/and remote users?

Learning paradigms. Most of the VMR-based learning and training activities in medicine require the development of experiential learning environments, whose design must be grounded on sound learning theories and approaches. The latter include the possibility to exploit serious games (SG) and gamification elements, which in turns require answering several research questions, such as: how to maximize the learning impact of SG? Which are the SG design model and framework most suited to achieve these objectives?

Validation studies. First of all, any use of technology within a medical scenario must be proven to work, particularly when patients are involved. Besides this clinical validation, it is necessary to implement other validation studies. As far as a learning tool is concerned, its effectiveness (in terms of learning outcomes and transfer to the professional context) must be validated. Finally, when dealing with VMR applications, it is clear that their usability and user experience must be carefully assessed as well.
Outline of work plan: During the first year, the candidate will survey the relevant literature related to the use of VMR in medial learning/training to (i) identify pros and cons of the available solutions, and (ii) develop a detail research plan to address such gaps. The literature survey will involve as well the fields of HCI and VMR collaborative environments, with specific focus on theoretical and design frameworks and best practices. In the first year, the candidate will conduct initial experiments related to the development of learning environment targeting individual trainees and (possibly) team activities.

During the second year and third years, the candidate will develop (through an iterative process of improvements and refinements) a set of solutions aimed at addressing the main research questions outlined in the proposal. In particular, design and development of these solutions will have a specific focus on (i) enabling effective collaboration among users during both training and learning, and (ii) validating the effectiveness of the proposed solutions along different dimensions (clinical, learning outcomes, usability/acceptability).
Expected target publications: The target publications will include virtual and mixed reality, serious games, learning technologies, and human computer interaction conferences and journals. Interdisciplinary conferences and journals related to medical training and learning will also be considered.
Possible journals include:
IEEE Transactions on Human-Machine Systems, IEEE Transactions on Visualization and Computer Graphics, ACM Transaction Human Computer Interaction,International Journal of Human–Computer Interaction, Journal of Computer-Supported Collaborative Learning, IEEE Transactions on Learning Technologies, IEEE Transactions on Emerging Topics in Computing, Computer and Education (Elsevier), Entertainment Computing (Elsevier) Relevant and well-reputed international conferences, such as: ACM SIGCHI, IEEE VR, IEEE IHCI, IEEE 3DUI, IEEE CIG.
Current funded projects of the proposer related to the proposal: None
Possibly involved industries/companies:The research program will be developed in strict cooperation with the SIMONVA center (Centro Interdipartimentale di Didattica Innovativa e di Simulazione in Medicina e Professioni Sanitarie) of the University of Oriental Piedmont, Novara, in the frame of a departmental convention with DAUIN (in which the proposer is the person in charge indicated by Politecnico). The director of SIMNOVA will co-tutor the candidate. SIMNOVA will contribute to the program with its expertise in high-level training, research and services in the health sector, with particular attention to the use of simulation as a tool for innovating training programs, improving the quality of care, reducing clinical risk and increasing safety for patients. Thus, SIMNOVA will take the role of domain expert in the definition of the requirements and of the design of the solutions implemented. SIMONVA will be also responsible of designing the clinical validation protocols and conducting the user validation of the developed tools.

Title: Biometrics in the wild
Proposer: Andrea Bottino
Group website: http://www.polito.it/cgvg
Summary of the proposal: Biometric traits are frequently used as an authentication system in a plethora of applications ranging from security to surveillance and forensic analysis. Today biometric recognition systems are accurate and cost-effective solutions. Thanks to these characteristics, they are starting to be deployed in a variety of scenarios, such as granting access to schools, health or leisure facilities, identifying patients in hospitals, developing pay-with-fingerprint systems and unlocking consumer devices like notebooks or smartphones. However, biometric recognition in the wild, that is biometric recognition from data captured in unconstrained settings, still represent a challenge and requires to face several issues. Among them, detection of region of interests (alignment, landmarking) in real-world data, segmentation of biometric traits, data normalization and fusion (at different levels) of multiple biometric traits. Another relevant issue in this context is the development of effective counter-spoofing measures. As a matter of facts, biometric recognition systems are vulnerable to more or less sophisticated forms of malicious attacks. The most common (and more simple for an intruder) type of attack is using fake biometric traits. Current counter-spoofing methods are based on the analysis of the liveness of the biometric traits. However, there is still a trade-off between security and accuracy, since introducing a spoof detector often causes a decrease of the acceptance ratio of the genuine traits without reducing to zero the false acceptance ratio.
Rsearch objectives and methods: The objective of this research proposal is to address several issues in the context of biometric recognition in the wild. In particular, the investigation will be focused on the following main topics.

Fusion approaches in the wild
Biometric recognition algorithm in the wild can exploit fusion approaches to improve their robustness. When focused on a single biometric trait (e.g., fingerprint, iris, face, voice), several features can be extracted from the incoming samples. These features represents different complementary “views” on the same data, each with its own peculiarity and drawbacks. For this reason, developing approaches that combine multiple features, capable of mutually exploiting their strengths and, at the same time, softening their weaknesses, could be a valuable solution to improve both the accuracy and the generalization properties of the classification system. A related approach refers to the use of multiple biometrics characteristics, used in conjunction to identify an individual. In both cases, tackling the fusion problem will require to:
- Investigate different feature extraction techniques, the relationship between different feature spaces and their confidence level for the task at hand; as far as it concerns biometrics in the wild, for some biometric traits, a necessary preliminary step for feature extraction will be also the development of robust techniques for segmenting in real-world data the region of interest where biometric traits are located
- Analyze the contribution of feature selection techniques, in order to avoid, during the integration of multiple feature sets, incorporating redundant, noisy or trivial information, which can seriously affect the performances of the recognition process
- Investigate different fusion approaches (e.g., early or late fusion) and their relevance to the classification problem

Liveness detection of biometric traits
One of the simplest form of attacks to a biometric recognition system is using fake biometric traits, which requires the development of specific protection methods capable of identifying live samples and rejecting fake ones. The objective of this task is the development of effective software based counter spoofing modules capable of rejecting fake samples without affecting in a sensible way the acceptance ratio of the live samples. This is particularly challenging when biometric recognition should be deployed in unconstrained environment, where the variability of the working conditions causes as well a dramatic decrease of the discriminability of the information obtained.

Optimization of the computational resources
One of the most relevant application area for biometrics in the wild is mobile biometrics (i.e., the use of biometric recognition systems on mobile/smart phones), which aims at guaranteeing the system robustness supporting as well portability and mobility, allowing its deployment in a wide range of operational environments from consumer applications to law enforcement. However, from a computational point of view, this requires as well the optimization of the resources required, in terms of computational power, memory requirements and algorithm complexity.
Outline of work plan: First year
Analysis of the state-of-the-art in the field of Biometrics in the wild. The expected outcome of this phase are (i) the identification of pros and cons of the current solutions and (ii) the preliminary design of methods capable of improving the available approaches.

Second year
The methods proposed in the first year will be thoroughly evaluated on publicly available benchmarks, which allow a comparison with a great variety of approaches in the literature. The expected results of this phase are the refinement of the proposed methods, and the design and evaluation on novel approaches on the specific domain.

Third year
The approaches analyzed during the first two years will be improved and possibly optimized in terms of resources required, in order to allow their deployment on a variety of different platforms.
Expected target publications: International peer-reviewed journals in the fields related to the current proposal, such as: IEEE Transactions Image Processing, IEEE Transactions on Information Forensics and Security, IEEE transactions Pattern Analysis and Machine Intelligence, Pattern Recognition, Pattern Recognition Letter, Image and Vision Computing, International Journal of Computer Vision. Relevant and well-reputed international conferences, such as: IEEE Face and Gesture Recognition (FG), IEEE International conference on Biometrics (ICB), Biometrics Theory, Applications and Systems (BTAS) conference, IEEE International Conference on Pattern Recognition
Current funded projects of the proposer related to the proposal: AStar project
Possibly involved industries/companies:No

Title: Programming Massively Parallel Server and Embedded Processors
Proposer: Stefano Quer
Group website: http://fmgroup.polito.it/quer/
Summary of the proposal: Mass-market computing systems that combine multi-core CPUs and many-core GPGPUs (General Purpose Graphical Processing Units) have brought terascale computing to laptops and desktops. Armed with such computing power, researchers have been able to achieve break-throughs in many disciplines using computational experiments that have been of unprecedented level of scale, controllability, and observability.
At the same time many new applications target embedded systems. These systems, pervasive in everyday life, are characterized by bounded computational resources on which the main target of any application is to maintain precision and reliability minimizing the computation effort.
Given this scenario, one of the major changes in the computer software industry has been the move from the serial programming paradigm to the multi-core and many-core parallel programming one.
Given the above considerations, this proposal concentrates on using new programming paradigms in different modern computation sciences mainly (but not only) related to automotive applications. Moreover, as many embedded applications can rely only on inexpensive hardware devices, the portability of several techniques to constrained systems represents a serious problem and it will also be analyzed.
Rsearch objectives and methods: GPGPUs are especially well-suited to address problems that can be expressed as data-parallel computations. In November 2006, NVIDIA introduced CUDA, a general purpose parallel computing architecture with a new parallel programming model and an instruction set that leverages the parallel compute engine in GPGPUs to solve many complex computational problems in a more efficient way than on a CPU.
Keeping these issues into consideration, one of the main target of this proposal is to achieve one of the following goals in automotive applications:
- To solve the same problem in less time.
- To solve bigger problems within the same given amount of time.
- To achieve better exact solutions for the given problem and the given amount of time.
- To approximate with more accuracy an exact solution with an estimated one using less resources or less time on resource-constrained architectures.

Among the high variety of modern cutting-edge automotive applications, we will more explicitly consider the following.

Motion planning and automated driving.
Path planning is the method for producing a continuous motion that connects a start configuration and a goal configuration while avoiding collision with known obstacles. One way to search for a path is by applying graph search algorithms. To do this, the configuration space must be converted into a graph, and eventually decomposed into subgraphs. In this scenario, not only the automated driving systems can be improved by becoming faster, more autonomous and avoiding human driver mistakes, but the computation and the energy efficiency may also be improved by optimizing the algorithmic design.
Moreover, obstacles are often described in a 2D or 3D maps while the path is described in the car’s configuration space. Many of the existing algorithms will have to include more accurate and finer-grained object recognition for different sort of objects (other vehicles, pedestrians, etc.) to improve user safety.

Cybersecurity and software inspection.
The increasing capability offered by connected computer systems enable a wide range of features and services, but with them comes the threat of malicious attacks. When the systems controlled are vehicles or vehicle related systems, the consequences of failure can be severe and the number of potential targets can be huge.
On the one hand, modern malware can automatically generate novel variants with the same malicious effects but appearing as completely different executable files. On the other one, the detection of network traffic anomalies or congestion due to successful intrusion can be detected by "deep packet inspection".
However, there is a tradeoff between the quality of the analysis and the power of the computing resources required. Machine learning techniques have been recently used within this domain and the same applies to many other cybersecurity related problems. Unfortunately these approaches are extremely expensive and they run much faster on GPGPUs than on CPUs. For those reasons many algorithms have to be redesigned and reimplemented on GPGPU architectures.

For our analysis, we will consider several platforms such as general and broadly accessible desktop and laptop GPGPUs, embedded mobile phone GPUs, and application specific computers such as the one within the Nvidia Drive PX series, which provide autonomous car and driver assistance functionality powered by other strategies, such as deep learning.
Outline of work plan: The work plan is structured in three years, as the PhD program.

In the first year the PhD student will improve his/her knowledge of writing parallel/concurrent software with advanced languages, mainly covering aspects not analyzed in standard curricula. The student will study the portability of standard algorithms on multi-core architectures, he/she will analyze data types and storage strategies, CPU to GPGPU communications, and he/she will better understand the scalability of an algorithm. The student will start to focus on the topics to analyze in the following two years, and he/she will also follow most of the teaching activity required by the PhD curriculum. Research activity will mainly target IEEE or ACM conference papers.

During the second year the work will be on designing and implementing new algorithms, or on optimizing known algorithms on new platforms, within one of the selected topics. Interdisciplinary aspects will also be considered. Research will start to target journal papers. Credits for the teaching activity will be finalized.

During the third year the activity carried forward during the second year will be consolidated, targeting new algorithms and optimizations.
Expected target publications: The target publications will be the main conferences and journals related to parallel computing, motion planning, robotics, sensors, and image processing. Given the nature of the project, interdisciplinary conferences and journals will also be considered.
Thus, the following represent some of the admissible targets:
- Elsevier Software: Practice and Experience
- Elsevier Concurrency and Computation: Practice and Experience
- MDPI Algorithms
- MDPI Sensors
- IEEE Transactions on Parallel and Distributed Systems
- IEEE Transaction of Intelligent Transportation Systems
- IEEE Transaction on Emerging Topics in Computing
- IEEE Access
Current funded projects of the proposer related to the proposal:
Possibly involved industries/companies:Possibly FCA and Magneti Marelli

Title: Concept Maps
Proposer: Laura Farinetti
Group website: http://elite.polito.it
Summary of the proposal: Concept Mapping was developed by Joseph Novak around 1972, based on Ausubel theory that meaningful learning only takes place when new concepts are connected to what is already known. Concept maps have been used in many different areas, including education, knowledge management, business and intelligence.
Currently, extensive Computer Science research is devoted to designing, creating and adapting algorithms for automatic concept maps extraction from resources, based on semantics to coherently arrange concepts and relationships. In this context data mining, semantic analysis, machine learning, named entity extraction and disambiguation are research areas that promise interesting results.
In the context of Learning Analytics, concept maps can be used as intellectual tools for many strategic tasks, for example (a) for the design and the assessment of educational resources, (b) for competence analysis, (c) for identifying competence gaps, (d) for assessing the relevance of learners’ generated content, (e) for improving learners’ awareness about their competence level, and about their educational target.

The technological research context for the PhD research is learning analytics, educational data mining, machine learning, semantic analysis, human computer interaction.
The application context is education, with the objective to design and create innovative solutions for teachers and learners, to improve learning effectiveness.
Rsearch objectives and methods: The PhD research will focus on the main technological challenges related to concept maps automatic extraction, validation, visualization, and on the design and implementation of innovative services and solutions that support learning and competence assessment. Specific emphasis will be given to concept map extraction for audio and video resources, with methodologies that encompass learning analytics, machine learning, semantic analysis, named entity recognition and disambiguation.

Specifically, research activities will address the following issues:
- Machine learning algorithms for named entity extraction and disambiguation, with special focus on audio and video content
- Concept maps automatic extraction and validation in different domains for different media resources
- Concept maps users’ friendly visualization strategies, adapted to different users’ profiles (teachers, learners, …)
- Concept maps comparison , for assessing competence gaps

Research results will be applied and validated in a number of innovative educational settings, with different objectives, such as:
- Assessing the relevance of educational resources in a specific learning scenario
- Designing educational settings that are coherent with a specific concept maps (curricula, resources, but also exam texts)
- Assessing the relevance of learners e-portfolios (where learners demonstrate their learning and development process in form of a collection of documents)
- Improving learners’ awareness and self-reflection on acquired competences, and competence gaps though concept maps visualization tools
- Personalize learning resources for specific students’ concept maps, or recommending learning resources and/or actions
Outline of work plan: Phase 1 (months 1-6): study of the state of the art in concept map automatic and semi-automatic extraction from multimedia resources: data mining and machine learning techniques; study of the state of the art in innovative applications using concept maps for education and for competence analysis.
Phase 2 (months 6-12): definition of the main objective of the research: identification of a challenging educational objective, identification of all the relevant educational resources, and identification of innovative technical strategies.
Phase 3 (months 12-24): design and implementation of the experimental setting: resource gathering; user modeling and profiling; concept map extraction, validation, visualization, comparison; competence analysis.
Phase 4 (months 24-36): complete demonstrator, user testing, reflections on learning effectiveness.
During all the phases of the research, the PhD candidate will have the chance to cooperate with other international academic institutions and with companies in the area of education and training, and to attend top quality conferences.
Expected target publications: IEEE Transactions on Learning Technologies
IEEE Transactions on Emerging Topics in Computing
IEEE Intelligent Systems
IEEE Transactions on Education
ACM Transactions on Computing Education (TOCE)
Expert Systems with Applications (Elsevier)
IEEE/ACM International Conferences on Learning Analytics, Data Mining, Learning Technologies (e.g., ACM LAK, ACM SIGCSE, IEEE COMPSAC, IEEE FIE), International Semantic Web Conference (ISWC).
Current funded projects of the proposer related to the proposal: No currently funded project.
An Erasmus+ (KA2, Strategic Partnership) proposal on this research area is currently under preparation.
Possibly involved industries/companies:Possible involvement: INDIRE - Ricerca per l’innovazione della scuola italiana.

Title: Approximate Computing for Power and Performance optimization
Proposer: Stefano Di Carlo
Group website: http://www.testgroup.polito.it
Summary of the proposal: Following the current trend, by 2040 computers will need more electricity than the world energy resources can generate. Internet-of-Things will soon connect 20 to 50 billion devices through wireless networks to the cloud.

This Ph.D. project focuses on the application of Approximate and Transprecision computing paradigms to optimize energy-accuracy trade-offs.

In many parts of the global data acquisition, transfer, computation and storage systems it is possible to trade off accuracy to either less power or higher performance. As examples, numerous sensors are measuring noisy or inexact inputs; the algorithms processing the acquired signals can be stochastic; the applications using the data may be satisfied with an “acceptable” accuracy instead of exact and absolutely correct results; the system may be resilient against occasional errors; and a coarse classification may be enough for a data mining system. By introducing accuracy as a new design dimension, the energy efficiency can even be improved by a factor of 10x-50x.

However, such large design space opportunity cannot be exploited without methods and tools able to quantify the impact of Approximate techniques on the accuracy of the system, allowing designers to take informed decisions. Addressing this problem is the main focus of this Ph.D. project.
Rsearch objectives and methods: In the last 10 years, the demand for new computing strategies driven by energy-efficiency has grown exponentially. Flop-per-watt (thus, per-euro) has become de-facto a driving model in hardware design. Results in this direction have been significant, leveraging first multi-core parallelism and then recently moving toward heterogeneous architectures (e.g., multicore CPU coupled with GP-GPUs). However, these evolutions will not be sufficient in the long term. To maintain an exponential increase in computational efficiency, we will need to rely either on an unlikely breakthrough discovery in hardware technology, or on a fundamental change in computing paradigms. The exploration of approximation in hardware and software from both a statistical and a deterministic viewpoint, as a computing paradigm shift to break the current performance and energy-efficiency barriers of systems at all scales, from sensors to supercomputers is at the base of the Approximate Computing (AxC) idea. However, AxC places formidable challenges across the entire computing software and hardware stack. Addressing these challenges requires balanced expertise in mathematics, algorithms, software, architecture design and emerging computing platforms.

In particular, the estimation of the impact of AxC techniques on the result of a computation, in most cases is based on running several times the computation using different configurations of AxC functions and operators (either hardware or software based). This process becomes computationally expensive, especially when a large set of design options makes the design space very large to explore.

The first tangible objective of this project is to explore a different paradigm by exploiting stochastic models to statistically evaluate the impact of the approximation without resorting to campaigns of application runs. The idea is based on the divide et impera paradigm. First, each candidate AxC technique is characterized as an isolated block in order to probabilistically model the error introduced by the approximation. Second, the knowledge of the error distribution of all considered operators is exploited to build a stochastic model of the full computation, thus modeling the approximation error propagation through the computation’s data flow.

The advantages of the proposed approach, compared to the state-of-the-art, are: (i) the characterization of the different operators that represents a complex task is done only one time creating a library of operators, (ii) the stochastic model of the computation can be automatically constructed by syntactically analyzing the application’s code without requiring a case by case analysis of the application, (iii) the evaluation of the model and the estimation of the effect of the approximation on the whole computation is fast and does not require several executions of the application.

The proposed model will not only consider the effect of the approximation on the precision of computation. It will be also linked with performance, power and reliability models in order to provide a full characterization of the approximated computation. Such a model opens a path for the development of a multi-objective design space exploration (DSE) framework able to explore the better options to apply AxC operators to a computation. This framework represents the second main objective of this Ph.D. project.
Outline of work plan: Year 1: cross-layer framework to model and simulate AxC applications.

The first step to achieve the goals described in the previous section is to create a complete modeling environment in which different cross-layer AxC operators can be characterized in terms of accuracy, power and performance impact, a large library of characterized AxC can be created and a complete AxC system architecture can be modeled and simulated at different abstraction levels.

To carry out this activity the student requires to acquire a deep knowledge of the state-of-the-art in order to create a taxonomy of the different AxC techniques and to select the candidates to include in the library. Moreover, the student must become familiar with several modeling, programming and simulation frameworks at different layers. The accuracy is not the only target of this framework, power, performance and reliability models must be also implemented.

Year 2: Fast and accurate prediction of the impact of AxC operators on a complex computation

The knowledge of the error distribution of all AxC operators considered in the library developed during the first year will be exploited as an input to build stochastic models of the full computation, thus modeling the approximation error propagation through the computation’s data flow.

Bayesian models are an example of a very efficient technique to model how stochastic events influence each other’s through complex interactions (which is the case of complex computations). Based on previous experience in the field from the research group, these models will be the first to be considered. However, the student will explore different option and will work to the development of new models customized for the specific problem.

With the construction of these models a system designer will be able to easily model his system (from hardware to software), using the developed framework, apply different AxC operators to different hardware/software portions of the system and assess the characteristics of the obtained system against the ones of a precise version.

Year 3: design space exploration

The last year of the Ph.D. will be dedicated to the development of DSE techniques for AxC systems. These approaches will exploit the capability of the stochastic models developed during the second year of the Ph.D. project coupled with efficient space exploration algorithms for multi-objective optimization such as the extremal optimization theory or other evolutive approaches. The use of DSE for multi-objective optimization will provide an instrument to trade-off accuracy with power and performance.

If time will be available, an attempt to include other design dimensions in the DSE process (i.e., reliability and security) will be done by exploiting results and expertise already available in the research group.
Expected target publications: The work developed within this project can be submitted to:

Conferences (DSN, DATE, DAC, IOLTS, etc.): a minimum of two conference publications per year.

Journals: IEEE TOC, IEEE TCAD, ACM TACO, IEEE TDSC: at least one publication during the second year and one during the third year.
Current funded projects of the proposer related to the proposal: The proposer submitted an ETN Marie Curie action on this specific topic
Possibly involved industries/companies:Intel Corporate Italia, Leonardo.

Title: Methods and Algorithms for Efficient Computational Physics-of-Matter of Electronic Devices in Harsh Environments
Proposer: Luca Sterpone
Group website: http://www.cad.polito.it
Summary of the proposal: Computational Physics is a growing and fundamental area of study, covering a vast list of topics focused to understand physical phenomena ranging from the soft matter to the dynamic computation of modern elaboration systems. When applied to Computer Machinery, the computation is viewed as a physical process with given characteristics, restrictions and limitations that arise from the natural world in which these computations are made. Technology and environmental considerations impose further constraints since computations are performed in VLSI integrated circuits. The relation of MOS device physics, the implications of scaling the feature size, the physics of the chip as a device itself, the implementation of software and the relation of the final dynamic computation with respect of the external harsh environment will be investigated in this PhD program. All these topics will be extrapolated considering the primary goal of developing a new generation of algorithms and software (based on parallel and reconfigurable computing paradigms) to model the interaction of physics particle into the VLSI matter of computing system during their functionalities. A robust layer of previously developed computational methods experimentally validated by radiation with high energetic particle experiments supports the PhD program.
Rsearch objectives and methods: There is a gap in today explanation of computer behavior in extreme environments. This gap consists of the unpredictability of computer dynamics when elaborating in harsh environments. Deep space, radioactive locations or even growing safety critical context such as autonomous driving are characterized by environments where the computational chips are invested by physical particles of different nature and energy that determine a variation of their functionality. This variation may require long time before becoming critical (cumulative dose effect) or affecting it suddenly in a transient or permanent way. The overall goal of this PhD program is to fill this gap providing a new and ground breaking solution to this problem.
The first objective of the PhD program will target the development of a new software capable to simulate the iterations of physical particles within VLSI basic cells structure of electronic devices and provide information about the cumulative effects such as total dose and displacement damage affecting them. The main innovation will be related to include the possibility to mimic the dynamic transition occurring in the cell during their functionality. A large set of particles will be considered: neutrons, heavy ions, protons and muons; as well as wide perspective of technology process will be analyzed toward ultra-nanometer devices (beyond 16nm). To achieve this goal, it is expected a drastic usage of parallel computing algorithm on Graphic Processing Units (GPUs) for two reasons: first of all, to provide a video evidence of the described phenomena and secondly, to speed-up the analysis.
The second objective of the PhD program will address the modeling of the computation during the affections of physical particles. While state-of-the-art approaches deeply provides static analysis, the correlation of physics particles into the software layer switching on the devices is still uncovered. The activity will be mainly based on Field Programmable Gate Arrays (FPGA) and it will be focused on the development of methods to correlate software algorithms implemented on various computational architecture with respect the physical behavior on the considered VLSI device.
GPUs, Digital Signal Processing (DSPs) and microprocessors will be used in order to provide a dynamic correlation with the particle physics irradiation and the software behavior.
The third and final objective of the PhD program will target the development of detection, correction and masking of multiple categories of particle-induced effects
(i.e., single event effects such as soft errors, single event upsets, single event transients, multiple events). All the methods will be developed considering the dynamic operation of the computational platform and considering their dynamic operation and marginal operating conditions such as low power, high speed.
The proposal will offer the opportunity to work on current, critical applications for deep space and radiation places as provided by the CERN center, which is supporting the experimental support of the PhD program. In details, CERN will provide the facilities where the experimental activities will be performed.
Outline of work plan: The activity is based on the following work packages:

WP1. (1 year) Computational Physics on dynamic VLSI Cell

This work package is devoted to the design of instruments and methods for simulating the particle interaction with electrons and molecules and by identifying clusters, delta rays, energy blobs and high-energy delta ray from secondary trajectory effects.
The work package will focus on both physical effects and VLSI layout characteristics with a particular emphasis on providing an effective particle effects-driven layout solution supported by specific software framework layers such as CAD tools and Layout cell masking in GDS-II format. The computational method developed will be based on parallel computing approaches and it will target dissemination on both computational physics and high performance computing conferences and journals.

WP2. (1st year/2nd year/3rd year) Harsh Computing

This work package will target radiation-induced effects on a wide range of ultra-nanometer devices: from FPGAs to Reconfigurable Computing Fabric (RCF) cores; while executing software benchmarks. The main target is the development of techniques able to fill-the-gap between the particle physics effect and the computational software. The techniques will take in account applicability of method related to standards such as DO-254 or ISO26262 and the respective SIL levels.

WP3. (3rd year) Particle physic in-circuit detector

This work package will have emphasis on the exploitation of the results achieved in the previous WPs and targeting the development of an in-circuit detector of particles capable to tune the physical characteristics of a cell in relation to the environment and to the VLSI structure. The techniques should principally address permanent faults such as the one induced by accumulated radiation (Total Ionizing Dose) or design-rupture such as Single Event Latch-Up (SEL).

WP4. (2nd year/3rd year) Beyond mitigation

Mitigations solutions are generally adopted at hardware and software level even if a large area and timing overhead is necessary. The WP4 will attack this paradigm trying to go beyond mitigation and towards charge sharing and particle shielding approaches which are able to intrinsically protect a computational system without the large area overhead of state-of-the-art mitigation approaches.

WPE (1st, 2nd and 3rd year) Experimental demonstrator

The existing contract active with the European Space Agency (ESA) and the consolidated experience with radiation facilities around Europe (CERN, UCL, GANIL and ISIS) and North America (TRIUMF and Los Alamos) will allow the PhD program to perform experimental analysis.
Expected target publications: Frontiers in Physics – Computational Physics
IEEE Transactions on Emerging Topics in Computing
Journal of Computational Physics
IEEE Transactions on VLSI
IEEE Transactions on Computers
IEEE Transactions on Reliability
IEEE Transactions on CAD
ACM Transactions on Reconfigurable Technology and Systems
Elsevier Computational Materials Science
Current funded projects of the proposer related to the proposal:
Possibly involved industries/companies:IROC Technologies, Cobham Gaisler, ESA, Thales Alenia Space, DLR, STMicroelectronics, NanoXplore.

Title: Cross-lingual text mining
Proposer: Luca Cagliero
Group website: https://dbdmg.polito.it
Summary of the proposal: Cross-lingual text mining consists of a set of methods and algorithms aimed at analyzing textual documents written in several languages. The aim is, on the one hand, to efficiently learn models portable to multi-lingual documents and, on the other hand, to effectively combine the information contained in documents written in different languages. Cross-lingual text mining finds applications in a number of research contexts, among which document summarization, topic discovery, social data mining, and text translation.

In recent years, many vector-based representations of words have been proposed (e.g., Word2Vect, GloVe, FastText). They analyze the context of use of a word to capture its semantic similarity with other words in the vocabulary. Although a significant research effort to train word embeddings from multi-lingual documents has been made, many text mining approaches are still not easily portable towards non-English documents.

The candidate will investigate new approaches to mining cross-lingual document collections. Specifically, she/he will address challenging cross-lingual text mining problems by exploiting the generality and expressiveness of vector-based word representations. Furthermore, she/he will test the applicability of the proposed approaches on documents written in multiple languages.
Rsearch objectives and methods: The research objectives address the following key issues.

- Alignment of multi-lingual word embeddings. The vector-based representations of the documents written in different languages need to be evaluated, compared with each other, and aligned in order to use them for cross-lingual text analyses.

- Study and development of new cross-lingual summarization algorithms. Summarization algorithms aim at extracting the most salient content from large document collections. The currently available summarization algorithms are able to cope with documents written in a limited number of languages. Furthermore, summaries are typically extracted from a set of documents written in the same language (i.e., the multi-lingual summarization task). To overcome these limitations, there is a need for developing new summarization approaches that are (i) portable to documents written in different languages, (ii) able to cope with cross-lingual collections, and (iii) scalable towards Big document collections.

- Topic discovery from cross-lingual collections. Discovering and characterizing the most salient topics in a set of documents is a well-known text mining problem. However, existing strategies mostly rely on language-dependent models. The aim is to extend existing methods to cope with cross-lingual collections. Advanced tools for topic discovery from cross-lingual collections (e.g., Reuters news, PubMed articles) will be developed.

- Heterogeneous document analysis. Collections of documents with heterogeneous characteristics often need to be analyzed. For example, social data include very short messages (e.g., tweets) as well as quite long posts and news articles. Furthermore, the vocabularies used in different types of documents are rather different. The goal is to explore the applicability of the proposed cross-lingual methods and algorithms on heterogeneous collections of documents acquired in different contexts.

Research methods include the study, development, and testing of various text mining algorithms. Extension of existing solutions tailored to specific types of documents (e.g. news articles) will be initially studied. Then, their portability to other contexts (e.g., social data) will be investigated.

To explore the portability of existing methodologies in cross-lingual domains, the performance of the models generated from large collections of documents written in many languages (e.g., Wikipedia documents, Reuters news) will be compared with that of language-specific solutions. The generated outcomes in English will be compared with English-written benchmarks (available in most application domains).
Outline of work plan: PHASE I (1st year): overview of existing word embedding models, study of their portability to different languages, and evaluation of different vector alignment strategies. Study of the state-of-the-art multi- and cross-lingual summarization algorithms, qualitative and quantitative assessment of the existing solutions, and preliminary proposals of new cross-lingual summarization strategies.

PHASE II (2nd year): study and development of new summarization algorithms, their experimental evaluation in a subset of application domains (e.g., e-learning, finance, social network analysis). Study of existing topic discovery techniques and analysis of their portability to documents written in different languages (e.g., Italian, French, Arabic).

PHASE III (3rd year): Study and development of new topic discovery techniques oriented to cross-lingual collections. Study of the portability of the designed solutions to heterogenous data (e.g., social data).

During all the three years the candidate will have the opportunity to attend top quality conferences, to collaborate with researchers from different countries, and to participate to competitions on data summarization organized by renowned entities, such as
- The contests of the MultiLing Community (e.g., http://multiling.iit.demokritos.gr/pages/view/1616/multiling-2017)
- the Data Challenges organized by the Financial Entity Identification and Information Integration (FEIII), and
- the Computational Linguistics Scientific Document Summarization Shared Task (CL-SciSumm) overseen by the National University of Singapore.
Expected target publications: Any of the following journals on Data Mining and Knowledge Discovery (KDD) or Data Analytics:

IEEE TKDE (Trans. on Knowledge and Data Engineering)
ACM TKDD (Trans. on Knowledge Discovery from Data)
ACM TIST (Trans. on Intelligent Systems and Technology)
IEEE TETC (Trans. on Emerging Topics in Computing)
IEEE TLT (Trans. on Learning Technologies)
ACM TOIS (Trans. on Information Systems)
Information Sciences (Elsevier)

IEEE/ACM International Conferences on Data mining and Data Analytics (e.g., IEEE ICDM, ACM SIGMOD, IEEE ICDE, ACM KDD, ACM SIGIR)
Current funded projects of the proposer related to the proposal:
Possibly involved industries/companies:

Title: Promoting Diversity in Evolutionary Algorithms
Proposer: Giovanni Squillero
Group website: http://www.cad.polito.it/
Summary of the proposal: Computational intelligence (CI) in general, and evolutionary computation (EC) in particular, is experiencing a peculiar moment. On the one hand, fewer and fewer scientific papers directly focus on EC as their main topic; on the other hand, EC techniques are routinely exploited “under the hood” in activities that are filed under the label “Machine Learning” — a quite attractive topic nowadays. Divergence of character, or, more precisely, the lack of it, is widely recognized as the most impairing single problem in the field of EC. While divergence of character is a cornerstone of natural evolution, in EC all candidate solutions eventually crowd the very same areas in the search space. Such a “lack of speciation” has been pointed out in the seminal work of Holland back in 1975, and it is usually labeled with the oxymoron “premature convergence”, that is, the tendency of an algorithm to convergence toward a point where it was not supposed to converge to in the first place. The research activity would tackle “diversity promotion”, that is either “increasing” or “preserving” diversity in an EC population, both from a practical and theoretical point of view.
Rsearch objectives and methods: The primary research objective is to analyze and develop highly-effective general-purpose methodologies able to promote diversity. That is, methodologies like the well-known “island-model”, that are not linked to a specific implementation nor to a specific paradigm, but are able to modify the whole evolutionary process.

This research started about a decade ago with “A novel methodology for diversity preservation in evolutionary algorithms” (G. Squillero, A. Tonda; GECCO Companion; 2008) and is more active every passing year. The proposer organized workshops on diversity promotion in 2016 and 2017; moreover, he presented tutorials on the subject at IEEE CEC 2014, PPSN 2016, and ACM GECCO 2018. This year, together with Dirk Sudholt, the research incorporated the theoretical runtime analyses of algorithms.

In more details, the study would start by examining why and how the divergence of character works in nature, and then trying to find analogies in the artificial environment. Such ideas would be first tested on reduced environment and problems used by scholars. Later, they will be generalized to a wide scenario, broadening their applicability for practitioners.

As “premature convergence” is probably the single most impairing problem in the industrial application of EC, any methodology able to ease it would have a tremendous impact. To this end, the proposed line of research is generic and deliberately un-focused, not to limit the applicability of the solutions. However, the research will explicitly consider domains where the proposer has some experience. Namely:
- CAD Applications, mostly related to the generation of Turing-complete assembly programs for test and validation of microprocessors.
- Evolutionary Machine Learning, that is mostly EC techniques used to complement traditional ML approaches.
- Computational Intelligence and games
Outline of work plan: The first phase of the project shall consist of an extensive experimental study of existing diversity preservation methods across various global optimization problems. General-purpose EAs developed in house will be used to study the influence of various methodologies and modifications on the population dynamics. Solutions that do not require the analysis of the internal structure of the individual (e.g., Cellular EAs, Deterministic Crowding, Hierarchical Fair Competition, Island Models, and Segregation) shall be considered. This study should allow the development of a, possibly new, effective methodology, able to generalize and coalesce most of the cited techniques.

During the first year, the candidate will take a course in Artificial Intelligence and Machine Learning, and all Ph.D. courses of the educational path on Data Science. Additionally, the candidate is required to learn the Python language.

Starting from the second year, the research activity shall include Turing-complete program generation. The candidate will move to MicroGP v4, a new toolkit currently under development. That would also ease the comparison with existing state-of-the-art toolkits, such as inspired and deap. The candidate will try to replicate the work of the first year on much more difficult genotype-level methodologies, such as Clearing, Diversifiers, Fitness Sharing, Restricted Tournament Selection, Sequential Niching, Standard Crowding, Tarpeian Method, and Two-level Diversity Selection. Publications targeting the “applications of computational intellifgence” are expected from this activity.

The theoretical aspects of the topic will be better analyzed together with Dr. Dirk Sudholt, from the University of Sheffield. The activity will be likely carried out through the COST project ImAppNIO. Joint publications are also expected from this cooperation.

At some point, probably toward the end of the second year, the new methodologies will be integrated into the Grammatical Evolution framework developed at the Machine Learning Lab of University of Trieste – GE allows a sharp distinction between phenotype, genotype and fitness, creating an unprecedented test bench, and G. Squillero is already collaborating with prof. E. Medvet on these topics. The final goal of this research will be to link phenotype-level methodologies to genotype measures.
Expected target publications: Journals with impact factors
- ASOC - Applied Soft Computing
- ECJ - Evolutionary Computation Journal
- GPem - Genetic Programming and Evolvable Machines
- Informatics and Computer Science Intelligent Systems Applications
- IS - Information Sciences
- NC - Natural Computing
- TCIAIG - IEEE Transactions on Computational Intelligence and AI in Games
- TEC - IEEE Transactions on Evolutionary Computation
Top conferences
- ACM GECCO - Genetic and Evolutionary Computation Conference
- IEEE CEC/WCCI - World Congress on Computational Intelligence
- PPSN - Parallel Problem Solving From Nature
Current funded projects of the proposer related to the proposal: The proposer is a member of the Management Committee of COST Action CA15140: Improving Applicability of Nature-Inspired Optimisation by Joining Theory and Practice
Possibly involved industries/companies:The CAD Group has a long record of successful applications of evolutionary algorithms in several different domains. For instance, the on-going collaboration with STMicroelectronics on test and validation of programmable devices, does exploit evolutionary algorithms and would definitely benefit from the research. Squillero is also in contact with industries that actively consider exploiting evolutionary machine-learning for enhancing their biological models, for instance, KRD (Czech Republic), Teregroup (Italy), and BioVal Process (France).

Title: Visual and Human-Centered Computing for Industry 4.0
Proposer: Fabrizio Lamberti
Group website: http://grains.polito.it
Summary of the proposal: There is a movement in the advanced economies which is referred to as “Industry 4.0” and aims to boost productivity and efficiency in industrial manufacturing by combining the latest advances in ICT.

Among technologies that are considered to be capable to play a key role there is “visual computing” that, in this context, can be defined as “the entire field of acquiring, analyzing, and synthesizing visual data by means of computers that provide relevant-to-the-field tools”. Together with approaches charactering the “human-centered computing” domain, visual computing broadly refers to computer graphics, computer vision technologies that can be used to capture, analyze, and interact with both the real and the virtual production worlds made up of connected, intelligent machines.

In the above scenario, this research will tackle some of the open problems in the field of visual and human-centered computing for industry, encompassing, e.g., the modeling of physical phenomena and their integration in interactive visualization systems, the implementation of “ecologic” training environments capable to replace real experiences, etc. Suitable strategies will be pursued to make the involved technologies become largely accessible and usable by industry’s end-users. To this aim, concrete proof-of-concepts will be developed by working on issues identified by relevant stakeholders.
Rsearch objectives and methods: Although computer graphics and computer vision have historically been used in a number of industrial applications, Industry 4.0 asks for a even more intensive exploitation of these technologies. Based on the above considerations, this research will be aimed to address some of the main application domains envisaged for visual and human-centered computing solutions, starting from open issues identified in the literatures as well as from concrete problems pinpointed by actual beneficiaries.

Based on activities carried out in the group in collaboration with important industries, a first field of investigation could be that of cyber-physical simulations. Since Industry 4.0 can be regarded as based on smart factories made up of connected, intelligent machines and systems, it appears clear that simulation of processes actuated by the above systems, both before and during operation, is a key aspect for achieving critical goals for production flexibility and efficiency. According to Industry 4.0 principles, simulation of relevant processes can be accompanied by the establishment of a strong connection between virtual and physical dimensions. In this scenario, aligned with the concept of “digital twins”, virtual visualizations may have to be seamlessly overlapped with both the physical objects which feed data in real time and the simulation model obeying the rule of physics, respecting mechanical constraints, etc. All of the above asks for computer graphics tools able to virtually reproduce the process in all its components, possibly in an interactive way, with the suitable level of detail. Simulations could benefit from immersive 3D visualizations, leveraging, e.g., Virtual and Augmented Reality-based environments supporting the integration of real-time factory data.

The second main focus could pertain the impact of human-machine interaction (HMI) on Industry 4.0. Indeed, HMI can be declined in many ways, e.g., considering user interfaces, ergonomics and usability factors, etc. In this perspective, the goal of this research will be to study how human tasks in the industry can be optimized, by considering the operation of machines and production lines. Attention will be concentrated on how intelligent, multimodal, natural interaction paradigms could be exploited to put the users in the center of production, incorporating them in the factory as knowledge consumers and producers with the aim to improve the performance and effectiveness of relevant processes.

Research carried out in the two domains above could be capitalized to address one of the major topics currently addressed by the group through the considered technologies, that is, computer-supported training. In fact, virtual simulations empowered with effective HMI solutions can allow industry end-users to engage in highly realistic training scenarios that would normally be too costly to implement (e.g., because of the need to interrupt normal operations), too dangerous to experience from real (e.g., like a wrong implementation of a security procedure, which may lead to injuries or fatalities), etc.

During the three years, other topics in the domain of visual and human-centered computing pertaining, e.g., information visualization and visual analytics, image processing and computer vision for industrial applications, etc. may be considered, based also on needs expressed by involved stakeholders.
Outline of work plan: During the first year, the PhD student will review the state of the art of technologies, methods and applications in the field of visual and human-centered computing and their applications to Industry 4.0, with the aim to identify up-to-date trends and most promising research directions by focusing on the main topics mentioned above and by possibly restricting the domain to relevant use cases linked, e.g., to funded projects by the proposer. Results of the analysis will be summarized in a review publication that will be submitted to a conference in the field. The student will complete his/her background on visual and human-centered computing as well as on related technologies by attending relevant courses.

During the second and third year, the PhD student will work on the design and development of one or more simulation frameworks that will enable the integration of data fed from industry machinery and processes, and will support the experimentation of vertical use cases targeting, e.g., digital-twin and training-oriented virtual experiences. Relevant research issues pertaining (the possibly simultaneous) interaction with real and virtual industrial elements will be duly considered. Once validated though both qualitative and quantitative observations gathered by directly involving end-users, results obtained will be summarized into at least another conference work plus a journal publication.
Expected target publications: International journals in areas related to the proposed research, including:
IEEE Transactions on Visualization and Computer Graphics
IEEE Computer Graphics and Applications
IEEE Transactions on Human-Machine Systems
IEE Transactions on Industrial Informatics
IEEE Transactions on Emerging Topics in Computing
ACM Transactions on Computer-Human Interaction
Relevant international conferences such as: ACM CHI, IEEEVR, GI, VIS, Eurographics, etc.
Current funded projects of the proposer related to the proposal: Topics addressed in the proposal are strongly related to those tackled in the Smart3D and VrRobotLine Regional projects managed by the proposer as well as to those of other projects/research contracts of the GRAphics and INtelligent Systems (GRAINS) group. Activities will be carried out within the laboratory established with the support of the “VR@Polito” initiative, co-funded by the Board of Governors of Politecnico di Torino.
Possibly involved industries/companies:KUKA Robotics, Brembo, Bell Production, SPEA

Title: Multivariate analysis and augmented reality visualization in research and industrial environments
Proposer: Bartolomeo Montrucchio
Group website: http://grains.polito.it/index.php
Summary of the proposal: Analysis and visualization of large quantities of data are becoming ubiquitous. Applications for finding and underlining links among data are used in many fields, in particular together with visualization tools, also in augmented reality.
This proposal is therefore addressed to the creation of new specific competences in this framework. The main case study will be by means of data provided by fiber bragg grating optical sensors developed and used in many applications by the PhotoNEXT POLITO Inter-Dipartimental Center.
The PhD candidate will be requested to use an interdisciplinary approach, with particular interest to operating systems, databases and of course image visualization techniques and augmented reality, possibly mixed with virtual reality.
Rsearch objectives and methods: Large quantities of data, such as those produced by sensors, in particular cameras, can be used to reveal hidden links among the same data. In order to do this, visualization is often the most important tool.
The main research objective is to analyze, visualize and propose to the user data coming from different kinds of sensors.
The first and most important case of study will be linked to the PhotoNEXT POLITO Center, in particular with relation to data coming from fiber bragg grating (FBG) optical sensors: they are sensors able to show quite little modifications in strain and temperature and can be used in several environments. At first a research environment will be taken into consideration. Then some case studies will be analyzed, in particular data coming from industrial and environmental applications, such as houses, bridges, airplanes, pipelines. For each of them specific visualization issues will arise, requiring software for data acquisition, databases, and visualization tools. Two examples could be a bridge and an airplane. For the bridge the final result will be to superimpose (in augmented reality) data from the FBG sensors to the bridge itself while e.g. a bus is passing. For the airplane (a drone) the visualization of the FBG sensors could be done in virtual reality over a CAD model of the plane during the flight. Augmented reality will therefore be investigated as a way to improve considerably fruition. Industrial applications will be carefully studied, also verifying the possibility of applying for patents.
The ideal candidate should have a good background in programming skills, mainly in C/C++ and should know well the main techniques of data storage and analysis, image visualization and augmented reality. Data coming from sensors (in particular FBG) must in fact be acquired and managed with an appropriate sampling frequency. Since interdisciplinary collaboration will be really important, in particular with physicists and experts in other fields of engineering, the ideal candidate should be interested in acquiring new competences and to work in interdisciplinary work teams.
Outline of work plan: The work plan is structured in the three years of the PhD program:
1- in the first year the PhD student should improve his/her knowledge of image visualization and data acquisition, mainly covering the aspects not seen in the previous curriculum; he/she should also follow in the first year most of the required courses in Politecnico. Since most of the work will be done in an interdisciplinary environment, he/she will be required to familiarize with the other components of the team, in order to be able to collaborate to the research and to scientific paper writing from the very beginning. At least one or two conference papers will be submitted during the first year. The conference works will eventually be presented from the PhD student himself.
2- In the second year the work will be both on designing and implementing new algorithms and on preparing a first work for a journal, together with another conference. Interdisciplinary aspects will be also considered. Credits for didactic will be also finalized.
3- In the third year the work will be completed with at least a publication in a selected journal. The participation to the preparation of proposals for funded projects will be required. If possible the candidate will be also required to participate to writing an international patent.
Expected target publications: The target publications will be the main conferences and journals related to image visualization, computer vision and image processing. Also interdisciplinary conferences and journals linked to the activities will be considered, in particular with relation to the industrial applications considered from PhotoNext POLITO Center. Journals (and conferences) will be selected mainly among those from IEEE, ACM, Elsevier, Springer and considering their indexes and coherence with 09/H1 sector. Examples are (journals):
IEEE Transactions on Visualization and Computer Graphics
IEEE Transactions on Industrial Informatics
ACM Transactions on Graphics
(conferences)
IEEE ICIP
ACM SIGGRAPH
Current funded projects of the proposer related to the proposal: The main collaboration of the proposer related to the proposal is with the POLITO Inter-Dipartimental Center for Photonic technologies (PhotoNext), which is focused on experimental and applied research in three key areas: optical fiber ultra-high speed communication systems, optical sensors and optical components for industrial applications. The interest here is linked to fiber bragg grating optical sensors, which produce a large quantity of data that require visualization.
Possibly involved industries/companies:Many different industries/companies could be involved in this proposal during the PhD period. It is important to note that often industries prefer patents to publications for industrial privative rights reasons. For this reason it will be important to consider, eventually, also patents, given current Italian National Scientific Qualification requirement.

Title: Human-level semantic interpretation of images and video
Proposer: Fabrizio Lamberti
Group website: http://grains.polito.it/
Summary of the proposal: Deep learning techniques such as deep convolutional neural networks (CNNs) have achieved impressive results in several computer vision tasks. However, solving complex tasks that involve high level understanding is still a challenging problem, and deep learning is often surprisingly brittle compared to human vision. There is increasing interest in endowing neural networks with better capabilities of modeling concepts and relationships, which is lacking in the standard CNN model.

Symbolic artificial intelligence encompasses a broad spectrum of paradigms based on high-level (human-readable) representations of knowledge, logic and reasoning. Their use in computer vision largely declined in the past decade since implicit representation learning has proven more efficient at elaborating complex sensory input and taking into account its inherent fuzziness. Nonetheless, symbolic knowledge representation excels precisely where statistical machine learning is weakest - in terms of transparency, ability to perform inference and reasoning, and representation of abstract concepts. Recent advances in neural-symbolic integration provide an avenue to combine the best of both worlds, namely the possibility of representing relationships between concepts with that of learning from data in an end-to-end fashion.

The candidate will study, design and evaluate novel methodologies for neural-symbolic integration for the semantic interpretation of images and video.
Rsearch objectives and methods: The overall objective of this research proposal is to explore novel ways to bridge the gap between deep representation learning and symbolic knowledge representation in computer vision, fueled by recent advances in the field of neural-symbolic integration. The candidate will research hierarchical models: at the lowest level, representation learning will be based on robust visual feature descriptors based on CNNs. At a higher semantic level, concepts and their relationships can be encoded and manipulated by employing knowledge representation techniques.

Specifically, this research proposal targets the area of neural-symbolic integration, which can be used to encode symbolic representation techniques, such as fuzzy logic, as tensors in a deep neural network, whose weights can then be learnt through relatively standard optimization technique. This formulation is particularly interest in computer vision as in principle, it allows to seamlessly insert and extract symbolic knowledge from a neural network, which can be used to impose strong prior. The potentiality of these techniques in the field of computer vision is still largely unexplored.

The candidate will target the application of neural-symbolic integration techniques to solve problems in semantic image interpretation, possibly expanding to video analysis and other complex datasets. There are a number of theoretical and practical issues to overcome to apply this approach at scale on complex visual tasks, and to make it easily applicable to the computer vision community. He/she will work on improving end-to-end joint training of both low level representation and high level symbolic level.

He/she will also progress towards the creation of visual knowledge bases, leveraging existing resources such as WordNet, Visual Genome and ConceptNet. Of particular interest is the possibility of guiding and improving the training process through imposing prior knowledge. Logical soft and hard constraints and relations can be specified compactly at the symbolic level: reasoning about such constraints can help improve learning, and learning from new data can revise such constraints thus modifying reasoning. One of the many challenges, however, is to identify data which is relevant to the visual domain.

During the PhD, new solutions will be studied and developed to address the issues listed above, and they will be evaluated and compared with state-of-the-art approaches to assess their performance and improvements. Experimental evaluation will be performed on realistic datasets and on available public datasets for benchmarking purposes.
Outline of work plan: During the first year, the candidate will build core competencies in deep learning / machine learning, symbolic representation, computer vision by attending PhD courses and seminars. He/she will survey the relevant literature to identify and compare existing frameworks and methods for neural-symbolic integration, based for instance on fuzzy first-order logic or Bayesian networks, and their current applications in computer vision. An evaluation of the most important research gaps is expected at this analysis. In the first year, the candidate will apply neural-symbolic integration techniques to specific computer vision tasks (e.g., object detection, action classification). An important intermediate goal is how to scale up existing methodologies to state-of-the-art very deep networks, that are likely required to solve non-trivial tasks. At least one conference paper will be submitted based on the work of the first year.

During the second year, the candidate will study advanced techniques for neural-symbolic integration and develop core reusable libraries. He/she will develop a visual knowledge base, based on existing resources such as WordNet, Visual Genome and ConceptNet, to derive prior knowledge that can be injected in the network to initialize training of the neural-symbolic architectures developed during the first year. Experimental comparison will be carried out on benchmark datasets to evaluate the effectiveness of injecting prior knowledge on training convergence and performance, and compare it with other strategies such as transfer learning. By the end of the second year, at least one paper in an international journal and one additional conference submissions are expected. The candidate will also achieve a firm grasp of the underlying theory and experimental methodologies for benchmarking machine learning models.

During the third year, the candidate will improve/refine the implemented methods, based on the results in the previous years; he/she will also test their robustness and generalizability in several, realistic datasets and tasks.
Expected target publications: The target publications will cover machine learning, computer vision and human computer interaction conferences and journals. Interdisciplinary conferences and journals related to specific applications (such as medical imaging) will also be considered. Journals include:

IEEE Transactions on Pattern Analysis and Machine Intelligence
International Journal of Computer Vision
IEEE Transactions on Human-Machine Systems
IEEE Transactions on Neural Networks and Learning Systems
Elsevier Pattern Recognition
Elsevier Computer Vision and Image Understanding
Current funded projects of the proposer related to the proposal: Private research grant, Reale Mutua Assicurazioni
Research project, Regione Piemonte, Smart3D
Possibly involved industries/companies:Leonardo
Reale Mutua Assicurazioni
A-star

Title: Intuitive machine learning for Smart Data applications
Proposer: Fabrizio Lamberti
Group website: http://grains.polito.it/
Summary of the proposal: Making machine learning models easier to train and understand is crucial to effective leverage Smart Data in a wide range of organizations. Techniques such as deep neural networks have shown unprecedented performance, but our understanding of their inner working is far from complete. Effective visualization techniques can facilitate the comprehension of machine learning models by practitioners and end users alike. For instance, several visualization techniques have been crucial to develop an intuitive understanding of how convolutional neural networks interpret images. This research proposal bridges theoretical and experimental analysis of machine learning techniques with the principles of scientific visualization and human-centered computing.

The candidate will design, implement and evaluate innovative approaches for visualizing and interacting with machine learning models and large scale feature representations. The research proposal intersects the following research areas:
(a) development of visualization and probing techniques that facilitate the understanding of deep learning models;
(b) unsupervised representation learning and visualization of data spaces;
(c) human-centered methodologies for interpretation, deployment and training of machine learning models.

The research activity will target application domains characterized by inherent data complexity such as image analysis, image captioning, medical imaging and exploration of large scale, multimodal datasets, in the context of ongoing projects.
Rsearch objectives and methods: The overall objective is to support effective and intuitive visualization of data models and machine learning applications by developing new general-purpose and application-specific tools.

The candidate will focus on research issues related to the visualization and understanding of deep neural networks and complex data spaces, including applications that challenge the current state of the art such as 3D images, videos and multimedia, medical data, and complex datasets. For instance, in the field of computer vision several visualization techniques (e.g., activation maps, occlusion studies, saliency maps) have been crucial to develop an intuitive understanding of convolutional neural networks.

The candidate will design effective, user-friendly and visual techniques to foster the understanding of machine learning models, possibly by engaging and interacting with the model. He/she will advance one or more of the following areas:
i) visualization and understanding of deep neural networks, with a focus on explainability; ii) techniques for representation and visualization of complex data spaces (such as deep embeddings, manifold learning, graph-based representations);
iii) developing techniques to “probe” complex machine learning methods; for instance, a few studies in literature have develop neural “probes” to experimentally determine the emerging properties of deep neural networks in the absence of a formal mathematical treatment.

The research plan will encompass both theoretical and experimental work on machine learning techniques, with principles of scientific visualization and human-centered interaction. The candidate will also study novel ways to support user interaction in data science, employing a wide range of interfaces (natural language interfaces, advanced visualization, immersive technologies, etc.). Interfaces targeted to different users (ML expert vs. end user) and applications (visual analytics, decision support systems, data exploration, visual query answering, etc.) will be developed. The goal is to allow users to derive insight into (or from) the model, as well as supply new information to the system thus supporting online/adaptive learning.

During the PhD, new solutions will be studied and developed to address the issues listed above, and they will be evaluated and compared with state-of-the-art approaches to assess their performance and improvements. Solid and reusable implementation of core visualization components is expected. Experimental evaluation will be conducted on real data collections. Validation will assess standalone performance as well as interaction with the user.
Outline of work plan: During year 1, the candidate will build core competencies in deep learning / machine learning, visual analytics, unsupervised representation learning and statistics by attending PhD courses and seminars. He/she will survey the relevant literature to identify advantages/disadvantages of available solutions. State-of-the-art visualization methods will be implemented/expanded for different applications domains, building a core library of reusable components. The candidate will submit at least one conference paper.

During year 2, the candidate will develop training/visualization and interaction methods for increasingly advanced models focusing on their interpretability and efficient user interaction. The algorithms will be tested on selected applications. By the end of year 2, at least one paper in an international journal and one additional conference submissions are expected. The candidate will also achieve a firm grasp of advanced deep learning / machine learning methods.

During year 3, the candidate will improve/refine the implemented methods, based on the results of previous literature, and focus on expanding user interaction capabilities. The candidate will design and carry out experiments testing the experience of users in realistic scenarios.
Expected target publications: The target publications will cover machine learning, computer vision, scientific visualization, and human computer interaction conferences and journals. Interdisciplinary conferences and journals related to specific applications (such as medical imaging) will also be considered. Journals include:

IEEE Transactions on Pattern Analysis and Machine Intelligence
International Journal of Computer Vision
IEEE Transactions on Human-Machine Systems
IEEE Transactions on Visualization and Computer Graphics
IEEE Transactions on Knowledge and Data Engineering
Elsevier Pattern Recognition
Elsevier Computer Vision and Image Understanding
IEEE Transactions on Medical Imaging
Medical Image Analysis
Current funded projects of the proposer related to the proposal: Private research grant, Reale Mutua Assicurazioni
Research project, Regione Piemonte, Smart3D
Possibly involved industries/companies:Leonardo
Reale Mutua Assicurazioni
A-star

Title: Development of Vulnerability-Tolerant Architectures
Proposer: Paolo Prinetto
Group website: http://www.consorzio-cini.it
Summary of the proposal: As with software, hardware vulnerability may result from project bugs or intentionally inserted faults (Hardware Trojans). In addition, unlike software, hardware can be observed and controlled (and therefore also physically attacked) from outside, through physical quantities and/or its physical interactions with the real world (Side-Channel effect).
In addition to the typical software attacks, aimed at the illegal theft of data and at the interruption of services, hardware is also subject to attacks aimed at the theft of the intellectual property associated to the technological solutions used, and to counterfeiting by means of fraudulent placement on the market of decommissioned and therefore typically worn out devices (Hardware Counterfeiting).
The PhD proposals aims at proposing, studying, and developing architectures capable of guaranteeing pre-defined security levels, even in the presence of vulnerabilities of varying nature, known and/or not yet revealed. These can be present in both hardware devices and/or in software applications.
The proposed solutions should be adaptable to the criticality of the target systems.
Rsearch objectives and methods: The research objectives will be:

1) Analyzing hardware security issues by considering different vulnerabilities.
2) Hardware vulnerabilities, regardless of their nature, can only be corrected by modifying the design and are therefore bound to remain permanently in the devices. To continue using vulnerable devices safely, it is necessary to develop architectural solutions capable of tolerating the vulnerabilities by preventing their exploitation by malicious attackers.
3) Several different solutions will be considered, depending on the type of vulnerability and the level of criticality of the system that uses the devices. These will include:
a. solutions based exclusively on software and aimed at preventing malicious attackers from exploiting known vulnerabilities;
b. where, in a way conceptually similar to smart cards, the affected devices are “confined” in protected (trusted) zones and are allowed to run only secure code, developed and loaded in protected and guaranteed environments, making it impossible to inject the malicious code that could be used to launch attacks of any kind;
c. solutions based on the interaction of appropriate combinations of different components with different characteristics, such as processors, FPGAs, Smart Cards, and dedicated hardware devices;
d. solutions aimed at tolerating behavior byzantine in the case of complex systems with many interacting devices [1].

While the results obtained during the research period are expected to be general and hold for any platform, the work during the PhD thesis will explicitly focus on the SEcubeTM platform ([2]), that will be made available to the candidate thank to a collaboration with Blu5 Labs and with the CINI Cybersecurity National Lab.
The SEcubeTM (Secure Environment cube) platform is an open source security- oriented hardware and software platform constructed with ease of integration and service-orientation in mind. Its hardware component is a SoC platform: a single-chip design embedding three main cores: a highly powerful processor, a Common Criteria certified smartcard, and a flexible FPGA.

[1] Leslie Lamport, Robert Shostak, Marshall Pease "The Byzantine Generals Problem" ACM Transactions on Programming Languages and Systems, Vol. 4, No. 3, July 1982, Pages 382-401.
[2] Varriale, Antonio, et al. "SEcube (TM): an Open Security Platform-General Approach and Strategies." International Conference on Security and Management (SAM). 2016.
Outline of work plan: The following steps will be taken:
1. Studying the actual implementation of real hardware architectures, also managing critical infrastructures, and identifying study cases;
2. Providing a Cyber Attacks classification for Hardware Trojans and Side-Channel attacks and studying how state-of-the-art identification and defense systems cope with them;
3. Performing a Vulnerability assessment of the selected study cases via Attack Campaigns on copies of portions of them within a custom dedicated Cyber Range;
4. Designing a set of architectural solutions to improve the security of the target study cases by defining vulnerability patterns and levels and possible countermeasures;
5. Emulating real systems from point 1 and integrating them with the above countermeasures techniques in scalable architectures, and performing a Vulnerability assessment of the same systems after having properly secured them with the proposed solutions;
6. Developing, comparing, and evaluating implementation for the different solutions and approaches identified in the research objectives, studying their pro’s and con’s;
7. Leveraging on the acquired experience, defining:
o a set of rules to be adopted to secure legacy hardware architectures;
o Design-for-Trust and Design-for-Security rules to be followed when designing new systems.

In the 1st year, the student is expected to perform steps 1-3.
In the 2nd year the student is expected to perform steps 4-5.
In the 3rd year the student is expected to perform steps 6-7.
The candidate will be tutored to fully reach the above-mentioned research objectives.
Expected target publications: Conferences:
- Design, Automation and Test in Europe (DATE)
- VLSI Test Symposium (VTS)
- HOST
- Specialized Workshops
- IEEE SandP
- ITASEC
Journals:
- IEEE Transactions on Dependable and Secure Computing
- IEEE Internet of Things
- ACM Transactions on Information and System Security
- ACM Transactions on Privacy and Security
- Design and Test of Computers
Current funded projects of the proposer related to the proposal:
Possibly involved industries/companies:Blu5 Lab
CINI Cybersecurity National Lab
Fondazione Links

Title: Improving the dependability and resilience to cyberattacks of next generation computer networks
Proposer: Riccardo Sisto
Group website: http://www.netgroup.polito.it
Summary of the proposal: The increasing pervasiveness of computer systems has led to the proliferation of distributed safety-critical systems connected to the internet. Smart Grids, autonomous driving vehicles, IoT smart devices, and Industry 4.0 are just examples of this trend. Ensuring dependability and resilience to cyberattacks of these distributed systems is as important as challenging.
The upcoming evolution of the networks towards 5G/IoT and the higher flexibility and dynamicity made possible by virtualization and Software-Defined Networking (SDN) bring new security threats and challenges, but at the same time offer new opportunities for protection, by enabling fast dynamic reconfiguration in response to cyberattacks. The main objective of the proposed research is to advance the state of the art in the techniques for exploiting this opportunity, by providing mechanisms for automated, fast, and provably correct policy-based reconfiguration of virtualized networks. Formal methods will be exploited to achieve correctness by construction, rather than a-posteriori verification. The main challenge is how to formally model next-generation networks, their components, and policies, in such a way that the models are accurate enough, and at the same time amenable to fully automated and computationally tractable approaches for ensuring safety and security by construction.
Rsearch objectives and methods: In the past, the main formal approach for ensuring the correct enforcement of network policies has been formal verification. According to this approach, a formal model of the network is built and analyzed, which can reveal the existence of potential network policy violations. A rich literature already exists about these techniques, even for SDN-based and NFV-based networks. The Netgroup has been active in this field in the last years. The formal verification approach has the disadvantage that, when a policy violation is found, a correction of the error (e.g. a change in the configuration of network components) has to be found manually. In rapidly evolving systems, such as the ones based on NFV and SDN, where network reconfigurations can be triggered frequently and automatically, manual steps are undesirable because they can limit the potential dynamics of these networks.
Specifically, the reconfiguration of a network to respond to a security attack should be totally automatic and very fast, in addition to being correct, not to disrupt legitimate traffic.
The candidate will explore a correctness-by-construction approach, alternative to formal verification, for ensuring the correct enforcement of safety and security policies of VNF- and SDN-based networks. The idea is to still use formal models of network components, as it is done with formal verification, but instead of just checking that the behavior of a fully defined model satisfies some policies (verification), some aspects of the model (e.g. the configurations of some network components) are left open, and a solution that assigns these open values is searched (formal correctness by construction). The aim of the research is to identify and experiment correctness-by construction approaches that exploit Satisfiability Modulo Theories (SMT) solvers. These tools are good candidates for this purpose because they are very efficient in determining if a set of logical formulas is satisfiable or not, and if it is, they can also efficiently find assignments of free variables that make the logical formulas true. The main idea to be explored in the research is to exploit an SMT solver as backend for building a tool that can automatically and quickly synthesize correct-by-construction configurations of NFV- and SDN-based networks that implement given policies. This implies finding ways of encoding the correct construction problem of such networks as an SMT problem that can be solved efficiently enough. The candidate will exploit the experience already gained by the Netgroup with SMT solvers for fast formal verification of NFV-based networks.
Till now, no researcher has yet tried this approach. The previously proposed solutions for automatic reconfiguration of networks generally use non-formal approaches or specially crafted algorithms and heuristics, and are limited to the configuration of just one type of function (e.g. firewall). If successful, this innovative approach can have high impact, because it can improve the level of automation in re-configuring next generation networks and at the same time provide the high assurance levels required by safety-critical distributed systems.
Outline of work plan: Phase 1 (1st year): the candidate will analyze the state-of-the-art of formal approaches for the modeling and verification of SDN-based, NFV-based, and IoT-based network infrastructures, with special emphasis on the assurance of security-related and safety-related network policies and with special attention to the approaches already developed within the NetGroup (Verigraph).
Also, recent literature will be searched extensively in order to find whether any new approach that goes in the direction of correctness by construction with formal guarantees (today not available) has appeared in the meanwhile.
Subsequently, with the guidance of the tutor, the candidate will start the identification and definition of the new approaches for automatic policy enforcement based on correctness by construction, as explained in the previous section. At the end of this phase, the publication of some preliminary results is expected. During the first year, the candidate will also acquire the background necessary for the research. This will be done by attending courses and by personal study.

Phase 2 (2nd year): the candidate will consolidate the proposed approaches, will fully implement them, and will make experiments with them, e.g. in order to study their scalability and generality. The approaches will be validated in the domain of industrial networks, by means of specific use cases, such as known attack scenarios. The results of this consolidated work will also be submitted for publication, aiming at least at a journal publication.

Phase 3 (3rd year): based on the results achieved in the previous phase, the proposed approach will be further refined and improved, in order to improve scalability, performance, and generality of the approach (e.g. number of different functions considered), and the related dissemination activity will be completed.
Expected target publications: The contributions produced by the proposed research can be published in conferences and journals belonging to the areas of networking (e.g. INFOCOM, ACM/IEEE Transactions on Networking, or IEEE Transactions on Network and service Management, Netsoft), security (e.g. IEEE SandP, ACM CCS, NDSS, ESORICS, IFIP SEC, DSN, ACM Transactions on Information and System Security, or IEEE Transactions on Secure and Dependable Computing), and applications (e.g. IEEE Transactions on Industrial Informatics or IEEE Transactions on Vehicular Technology)
Current funded projects of the proposer related to the proposal: ASTRID (AddreSing ThReats for virtualIseD services) H2020 project.
Possibly involved industries/companies:TIM

Title: Blockchain for Industry 4.0
Proposer: Fabrizio Lamberti
Group website: http://grains.polito.it/
Summary of the proposal: A blockchain is a distributed ledger maintained by network nodes, which records transactions executed between nodes (in the form of messages sent from one node to another). Information inserted in the blockchain is public, and cannot be modified or erased. Smart contracts are self-executing contracts (generally saved on a blockchain) whose terms are directly written into lines of code.
Among other fields, Industry 4.0 could largely benefit from the blockchain technology. For instance, blockchain could be used to implement a secure authentication and communication layer for machines, or to provide a shared ledger for tracing all the events occurred during production. Similarly, smart contracts could be used to let machines autonomously interact, based on a set of immutable, hard-coded rules.
The present Ph.D. proposal aims at improving the state of art of research concerning blockchain technology in Industry 4.0. The Ph.D. candidate will explore existing blockchain and Distributed Ledger Technology (DLT) frameworks in order to identify the most suitable ones to be used in an Industry 4.0 context. He/She will then propose and develop blockchain-based solution(s) to enable the intelligent interaction of machines, sensors and other devices used, e.g., in production, maintenance and logistics.
Rsearch objectives and methods: The Industry 4.0 scenario currently presents several issues/challenges, such as, among others:
- the need to enable intelligent decision-making and negotiation mechanisms, e.g., letting smart machines autonomously take decisions and negotiate with each other;
- the need to continuously monitor the production as well as machines' status and operation history, e.g., to identify malfunctions or bottlenecks;
- the need to foster cyber and property security, e.g., to grant the access to private information or the control of physical machines only to authorized entities.

The proposed Ph.D. proposal aims to address the above points by leveraging blockchain/DLT technology. In particular, three objectives will be pursued, as reported below.

- Evaluation of existing blockchain/DLT frameworks with respect to the requirements of Industry 4.0: to reach this objective, the Ph.D. candidate will perform an extensive analysis of the state of the art and will identify the main advantages and disadvantages of each framework (e.g., in terms of transactions/second, smart contracts support, infrastructure costs, etc.). During this activity, he/she might also define taxonomies enabling the comparison of the above frameworks.
- Identification of innovative use cases in Industry 4.0, which could exploit blockchain/DLT to improve automation: to reach this objective, the Ph.D. candidate will investigate existing Industry 4.0 processes/applications/frameworks, with the objective to identify the ones could benefit the most from blockchain/DLT. In this phase, state-of-art works will be considered to underline current issues characterizing Industry 4.0 processes. The Ph.D. candidate may dialogue with companies working in the Industry 4.0 field (possibly collaborating already with the GRAINS research group), to elicit their needs. At the end of this phase, a list of use cases will be drafted. In parallel, the Ph.D. candidate will perform a review of blockchain/DLT existing/proposed applications in Industry 4.0 (by focusing both on innovative solutions proposed in literature or by startups). Based on the findings of such review, the list will be revised to identify, among selected use cases, the most innovative ones not yet explored/implemented. Selection may be performed by involving the companies mentioned above.
- Development of blockchain/DLT-based application(s) for Industry 4.0: to reach this objective, the Ph.D candidate,will select, based on the identified use case(s), the most suitable blockchain framework/DLT and will acquire the required (programming) competencies to develop one or more applications. Selection will consider, among others, the availability of blockchain infrastructures, at Politecnico or at companies' facilities (especially for permissioned blockchains), as well as requirements on privacy, costs, transactions/second, etc.

As previously stated, some of the activities mentioned above may be performed in cooperation with companies working in the Industry 4.0 field, or having previous experience with blockchain/DLT technology. In this perspective, it is worth underlining that the GRAINS group is currently involved in a number of research activities related, among others, to supply chain and additive manufacturing. The GRAINS group already performed preliminary investigations of blockchain/DLT technology in use cases related to these fields. The Ph.D candidate could continue such investigations, or focus on alternative/innovative use cases.
Outline of work plan: Phase 1 (1st year): acquisition of knowledge, skills and competencies related to blockchain/DLT technology. In this phase, the Ph.D. candidate will perform a literature review by focusing on technical aspects of existing blockchain/DLT frameworks. He/She could then define a taxonomy to enable their comparison according to several dimensions, ranging from technical aspects (such as transactions/second), privacy features, costs, etc. The outcome of this phase will be a publication in the proceedings of an international conference or in an international journal.

Phase 2 (2nd year): acquisition of knowledge, skills and competencies related to Industry 4.0. In this phase, the Ph.D. candidate will deepen his/her knowledge concerning technology related to Internet of Things, Internet of Services, cyber-physical systems, additive manufacturing, etc. and will study current industrial processes (also in collaboration with companies in the field). This phase will enable him/her to identify some limitations of the current processes, which could be overcame by relying on blockchain/DLT. In parallel, he/she will investigate which applications leveraging blockchain/DLT have been already proposed in the Industry 4.0 scenario, in order to pinpoint the most innovative use cases to focus on. The outcome of this phase will be a publication in an international journal. Based on the selected use case(s) the Ph.D. candidate will also identify one or more blockchain/DLT frameworks, and will acquire the needed (programming) competencies.

Phase 3 (3rd year): development of blockchain/DLT-based application/s. This phase will start with an analysis of requirements (eventually carried out with the collaboration of involved companies), and will result in the definition of the architecture of the application/s and in its/their development and testing. The end of the year will be devoted to writing one or more journal papers describing activities and obtained results.
Expected target publications: IEEE Transactions on Industrial Informatics
IEEE Transactions on Emerging Topics in Computing
IEEE Internet of Things
IEEE Access
IEEE IT Professional Magazine
IEEE Industrial Electronics Magazine
Industrial Management and Data Systems Journal
Computers in Industry Journal
IEEE International Conference on Industrial Engineering and Engineering Management
IEEE International Conference on Industrial Informatics
IEEE International Conference on Emerging Technology and Factory Automation
IEEE International Conference on Automation and Computing
European Conference on Smart Objects, Systems and Technologies
Current funded projects of the proposer related to the proposal: General Motors – private research grant "Disrupting vehicle controls and fleet management: block-chain, parallel computing and quantum computing"
General Motors - private research grant "Connected Manufacturing project"
Possibly involved industries/companies:

Title: Effects of Data Quality on Software Application Bias and Mitigation Strategies, Techniques and Tools
Proposer: Marco Torchiano
Group website: https://softeng.polito.it
Summary of the proposal: In computer science, “garbage in, garbage out” (GIGO) is a popular idiom to identify cases where “flawed, or nonsensical input data produce nonsensical output”. The GIGO principle implies that the quality of a software service of application is sensibly affected by the quality of the underlying data.

Nowadays, many software systems make use of large amount of data (often, personal) to make recommendations or decisions that affect our daily lives. Consequently computer-generated recommendations or decisions might be affected by poor input data quality. This implies relevant ethical considerations on the impact (in terms of relevance and scale) on the life of persons affected by the output of software systems.

In addition to the above aspects of data quality and biases, the software itself may suffer of poor quality too and its algorithms might implicitly incorporate bias in their own logic: the well-known trolley problem is an example of this issue.

The PhD proposal aims at investigating the impact of poor data quality and biases in the data on the automatic decisions made by software applications. As a minor aspect, the ethical character of algorithms and the relative effects on decisions will be also investigated.
Rsearch objectives and methods: The objectives of the PhD plan are the following ones:
- O1: Build a conceptual and operational data measurement framework for identifying data input characteristics that potentially affect the risks of wrong or discriminating software decisions. This goal encompasses identifying which characteristics have an impact, and the measurement procedure.
- O2: Collect empirical evidence concerning the actual impact of the measured data quality issues on automated decisions made by software systems. The evidence will be built by means of different research methods: case studies, experiments or simulations, depending on the availability of data, software and third-party collaborations. In particular a key achievement is the establishment of relational links between quality issues and output bias features.
- O3: Design of mitigation and remediation strategies and specific techniques to reduce the problem. A proof of concept implementation should be provided. We anticipate not all aspects of the problem will be solvable computationally, in such cases it will be important to advance the knowledge in the area by identify explanations and provide critical reflections.

In addition, as a secondary goal:
- O4: Investigate how quality of software and bias incorporated in the algorithms can contribute to flawed decisions made by software applications;
o O4.1: If any evidence is found, investigate how this aspect is related to the previous one of data quality and bias
o O4.2: Design and prototyping of remediation techniques for the problem.
Outline of work plan: The Ph.D. student will work within the Softeng research group and in strict collaboration with Nexa Center for Internet and Society, interacting with a research multidisciplinary environment and collaborating with computational social scientists, machine learning researchers, network engineers on one side, and scholars in philosophy, law, and social sciences on the other side. The student will also interact with the collaborators listed in the next field of the template.

The first year will be devoted to the definition of the conceptual and operational data measurement framework. The foundations of the framework will be the standards ISO/IEC 25012 (Data quality model) and ISO/IEC 25024 (Data quality measurement), together with relevant initial literature identified by the supervisors.

Approximately from half of the first year to the end of the second year the student shall design and conduct empirical investigations to achieve the goals above defined. Visiting periods abroad might be considered.

The third year will be mainly dedicated to design and prototyping technical remediations to the problems found, when feasible; otherwise to provide critical reflections on the issues.

The thesis shall also include lessons learned, improvements and future work based on the evaluation results of the previous years.
Expected target publications: Illustrative examples of targeted scholarly journals include:

IEEE Computer
Transactions on Software Engineering and Methodology
Government Information Quarterly
Journal of Systems and Software
Software Quality Journal
Information
IT Professional
IEEE Software
Philosophy and Technology
Big Data and Society
Communications ACM
ACM Transactions on Information Systems
Significance
Daedalus
Information sciences
Engineering Applications of Artificial Intelligence
Nature machine intelligence
Current funded projects of the proposer related to the proposal:
Possibly involved industries/companies:UNINFO – Technical Commission 504 (Software Engineering)
ISO/IEC Joint Technical Commission 1 (Software Engineering)
TIM

Title: Efficient Functional Model-Driven Networking
Proposer: Fulvio Risso
Group website: http://netgroup.polito.it
Summary of the proposal: This research activity aims at redefining the networking domain by proposing a functional and model-driven approach, in which novel declarative languages are used to define the desired behavior of the network. Furthermore, networking primitives are organized in a set of inter-changeable functional modules that can be flexibly rearranged and composed to implement the requested services.
The current proposal will allow network operators to focus on what to do, instead of how to do, and it will also facilitate the operations of automatic high-level global orchestration systems, which can more easily control the creation and update of network services. However, the high-level view guaranteed by the functional approach should not give up to the efficiency of the implementation, will be taken into high consideration.
The prototype implementation will target the Linux operating system through the eBPF component, which will be leveraged to provide networking services for major open-source orchestration frameworks such as Kubernetes and OpenStack.
Rsearch objectives and methods: The objective of this research is to explore the potential of a functional and model-driven approach to create, update and modify network services, without affecting the efficiency. Such an approach allows (i) to deploy complex network services by means of functional languages, i.e., focusing on the behavior that has to be obtained (also known as intent-driven approach); (ii) to create elementary functional components that can provide basic network functionalities and that can be arbitrarily rearranged and composed in complex service chains in order to deliver the expected service; (iii) to deploy optimized version of each components based on run-time gathered information, such as the actual traffic patterns, the availability of resources (e.g., CPU, memory, hardware coprocessors such as SmartNICs).
The above objective will be reached by investigating the following areas.

Functional languages for complex service composition and service description. Novel languages are required to specify the functional behavior of the network, starting from existing research proposals (e.g., FRENETIC, http://www.frenetic-lang.org/; group-based policies, http://gbp.readthedocs.io/ ) and extending the above approaches to higher-level modular constructs that facilitate compositional reasoning for complex and flexible service composition. This requires a functional modelling or each network function as well, which enables the creation of abstract functions, possibly with multiple implementations. This implies the definition of a functional model for each elementary network module that can define exactly which incoming network traffic is accepted, how it is possibly modified, and what will be returned in output. This model, which resembles to a transfer function, enables the creation of complex network models operating at the functional level, hence providing a match between user requests, coming from service-wide functional languages, into the delivered network service. In a nutshell, the requested service will be obtained by the composition of the above elementary (functional) blocks, and the overall service properties are obtained thanks to the formal composition properties of each transfer function.

Model-driven creation of efficient elementary network components. High-level languages are definitely important as they represent the entry point of potential users (e.g., network operators) in this new intent-based world- However, without an efficient mapping on the underlying infrastructure, they may never be adopted by potential users. This second direction of investigation aims at creating modular networking building blocks that provide elementary functions, with the capability to be flexibly rearranged in complex service graphs defined at run-time. A model-driven approach brings additional flexibility in the system as each component can be selected based on what it does instead of the actual implementation, enabling massive reuse and improvement of components without affecting the functional view of the service. A possible starting point for achieving efficiency consists in leveraging some existing technologies, such as the eBPF virtual CPU (https://cilium.readthedocs.io/en/latest/bpf/ ), which propose networking-tailored micro virtual machines running in the Linux kernel, and the model-driven development of networking components proposed in many communities (eg., OpenDayLight, ONOS, NetConf/YANG). Particularly, the former will be exploited to guarantee the efficient implementation of the above services, thanks to the capability to modify the running code at run-time and adapt the software the surrounding execution environment (e.g., traffic patterns, usage of resources).
Outline of work plan: Modelling of Virtual Network Functions (Months 1-12): this activity involves the definition of a functional model that captures the characteristics of (at least) the most common network functions. This enables the creation of functional blocks that may have different implementations, which can be possibly chosen at run-time based on additional constraint (e.g., number of users, requirement of resources, possible dependencies such as availability of a given hardware coprocessor). This work can be possibly carried out with the collaboration of prof. Sisto. Depending on the outcome, this work will be submitted for publication either in a major conference or in a journal.

Functional languages for complex service composition (Months 13-18): this activity involves the definition of a functional language that enables the creation of complex services by means of declarative constructs. A prototypal compiler will be developed that translates high-level constructs into actual service graphs, exploiting he functional modules defined above. This work will be submitted for publication in a major conference.

Model-driven creation of efficient elementary network components (Months 19-30): this will represents the major part of the work and it consists in two logically separated steps. First, the model-driven creation of components that involves mainly the control and management plane, which need to be generic enough to be suitable for any generic service of a given type (e.g., firewall). Second, an efficient implementation of a subset of the above services leveraging the eBPF and the recently added Express Data Path (XDP) technologies, both available in the Linux kernel. This is expected to originate at least three publications, the first presenting the model-driven approach to modular networking component; the second focusing on creation of efficient components; the third that includes an extension of both previous topics and that will be submitted to a major journal.

Datacenter-wide orchestration (Months 31-36): this phase will conclude the project with the definition of a set of orchestration algorithms that are able to orchestrate a service across the entire datacenter. This will be translated into a proof-of-concept orchestrator that creates a complex network service given a set of high-level requirements and constraints, showing the advantages of the proposed technologies. Particularly, this task may leverage the experience of the ASTRID project, which focuses on orchestration of network security services. This work is expected to be submitted to a major journal.
Expected target publications: Most important conferences:
- USENIX Symposium on Networked Systems Design and Implementation (NSDI)
- USENIX/ACM Symposium on Operating Systems Design and Implementation (OSDI)
- IEEE International Conference on Computer Communications (Infocom)
- ACM workshop on Hot Topics in Networks (HotNets)
- ACM Conference of the Special Interest Group on Data Communication (SIGCOMM)

Most significant journals:
- IEEE/ACM Transactions on Networking
- IEEE Transactions on Computers
- ACM Transactions on Computer Systems
- Elsevier Computer Networks
Current funded projects of the proposer related to the proposal: Grant from Huawei Research (Futurwei)
H2020 European project ASTRID
Possibly involved industries/companies:Huawei

Title: Machine learning for sentiment analysis
Proposer: Elena Baralis
Group website: http://dbdmg.polito.it/
Summary of the proposal: Sentiment analysis is the process of converting unstructured data (typically text) to extract the attitude of the creator of the content in that specific context. This process helps converting “sentiments” into a machine-understandable format, which in turn enables a series of possibilities, e.g., understand a crowd’s opinion about a specific topic, measure customer satisfaction, and offer special care for those individuals that are having a negative experience.

Despite the advances in the field, sentiment analysis still struggles in some aspects (e.g., sarcasm detection).
The availability of further content types (e.g., audio, video) may significantly improve the performance of current techniques. Hence, a key issue in sentiment analysis will be the capability to design machine learning techniques capable to deal with heterogeneous contents and complex data relationships.

Ongoing collaborations with the SmartData@PoliTO research center, universities (e.g., Eurecom) and companies (e.g., RAI, GM Powertrain, ENEL, ENI) will allow the candidate to work in a stimulating international environment.
Rsearch objectives and methods: The objective of the research activity is the definition of novel sentiment analysis approaches aiming at improving the detection performance by considering heterogeneous information sources.

The following steps (and milestones) are envisioned. Data collection and exploration. Publicly available datasets will be initially considered as benchmarks. Subsequently, new data will be collected (e.g., by means of twitter APIs) with the aim of considering different data formats and types (e.g., images, emoji). Tools for explorative analysis will be exploited to characterize data and drive the following analysis tasks.

Sentiment analysis algorithms design and development. Novel algorithms designed for the specific data analysis problem will be designed. The algorithms will exploit the content derived from different information sources (e.g., by performing object detection in images) to extend the context provided by textual information. Graph representations will be considered to represent correlated (positively or negatively) concepts, in terms of sentiment expressed by the different users.

Deployment in real world applications. A variety of different application scenarios will be considered as targets for the proposed sentiment analysis techniques, e.g. recommender systems, reviews management, and creation of content tailored to a specific audience (in the marketing domain). Big data platforms will be considered as possible development frameworks, given the large data volumes to be considered.
Outline of work plan: PHASE I (1st year): state-of-the-art survey for algorithms and available platforms for sentiment analysis and textual data analysis, assessment of main sentiment analysis techniques on several case studies (e.g., Twitter data, Amazon product data); performance analysis and preliminary proposals of optimization techniques for state-of-the-art algorithms; exploratory analysis of novel, creative solutions for considering new data types.
PHASE II (2nd year): design of new algorithms, gradually introducing different data types; development, experimental evaluation on public datasets of increasing complexity; implementation on a subset of selected big data platforms.
PHASE III (3rd year): algorithms improvements, both in design and development, experimental evaluation in new application domains, e.g. recommendation systems.

During the second-third year, the candidate will have the opportunity to spend a period of time abroad in a leading research center.
Expected target publications: Any of the following journals
ACM TOIS (Trans. on Information Systems)
IEEE TKDE (Trans. on Knowledge and Data Engineering)
ACM TODS (Trans. on Database Systems)
ACM TKDD (Trans. on Knowledge Discovery in Data)
Journal of Big Data (Springer)
IEEE TBD (Trans. on Big Data)
ACM TOIT (Trans. On Internet Technology)

IEEE/ACM International Conferences (e.g., IEEE ICDM, ACM KDD, IEEE BigData, IEEE Asonam)
Current funded projects of the proposer related to the proposal: I-React (Improving Resilience to Emergencies through Advanced Cyber Technologies) - H2020 EU project - http://www.i-react.eu/
Possibly involved industries/companies:

Title: Automotive SoC Reliability and Testing
Proposer: Paolo Bernardi
Group website: http://www.cad.polito.it
Summary of the proposal: Nowadays, it is continuously growing the number of integrated circuits included in critical environments such as the automotive field. For this reason, semiconductor manufacturer has to guarantee the reliability of the released components for the entire life-cycle that can be up to 10 -15 years.

The activities planned for this proposal includes efforts towards
- The optimization of crucial reliability measurements such as Burn In and Repair strategies
- The design of test strategies aimed at supporting error detection in-field which are demanded by recent standards such as the ISO 26262
- The application of machine learning methodologies to elaborate diagnostic manufacturing volume result.

The project will enable a phd student to work on a very timely subject and supported by companies currently collaborating with the research group.

The potential outcome of the project is both related to industrial advances and high quality publication.
Rsearch objectives and methods: The phd student will pursue the following objectives in his research activities. The major subjects are numbered 1 to 5, while more specific objectives for every major point are listed as bullet sub-lists.

Reliability measurement: in this field, the student will progress the state of the art in the following topic:
1. Improve the effectiveness of TEST and STRESS pattern generation through thermal analysis and modeling
- Better coverage of defects such as delay and exacerbation of intermittent faults
- Burn-In reduction time by optimized test and stress sequences at CPU and System-on-Chip levels.
2. Introduction of innovative flows to increase the yield in the manufacturing process
- Novel repair algorithms for embedded cores, including replication of modules and alternative functionality adoption in case of identified failures
- False positive classification of devices analysis and its reduction along volume data analysis

In-field testing: in this context, the improvement with respect to current literature and industrial practice is related to the following topics. This is a very timely subject also strongly connected to industrial requirements.
3. Development of methods for achieving high coverage of defects appearing along mission behavior, as demanded by the ISO 26262
- Key-on and runtime execution of Software-based Self-Test procedure for CPUs and peripheral cores
- Diagnosis trace for faster failure analysis of returns from fields.
- Implementation of Infrastructure-IP and redundancy methodologies for supporting effective and efficient on-line self-test

Machine learning methodologies applied to test: for this last subject, the student will explore the concept of using Machine Learning strategies applied to manufacturing and on-line testing.
4. Conception and implementation of machine learning methodologies to elaborate diagnostic manufacturing volume result
5. Exploration of on-chip Artificial Intelligence adoption in Functional Safety.
Outline of work plan: With respect to the main topics of investigation, the list of specific actions per year are listed below. The activity plan for every year includes actions related to the 3 main subject of investigation (please see the previous section).

1st year: in the first year, the phd student will work on the following Reliability Measurement techniques and In-Field Testing methods.
a) Repair strategies overview for SoC components
b) Memory Error Correction Codes (ECC) and Repair strategies
c) Design and realization of Database suitable to store the downloaded information and whose structure suits for a large number of measure type and temporal information
d) SoC sensitivity evaluation, mission mode Failure in Time (FIT) measurement and estimation

2nd year: during the second year the student will progress and refine the following Reliability Measurement subjects, improve capabilities of in-field testing techniques developed during the first year; machine learning techniques will be developed that progress the state of the art.
a) Memory ECC and repair strategies (continuation of the first year)
b) Software-assisted Built-In Self-Test methods: design and implementation
c) Diagnosis strategies development for quickly feedback failures appearing along Burn In or causing field returns
d) Volume data analysis and implementation of Machine Learning to improve memory testing capabilities.

3rd year: during the third year, the student will further refine the implemented techniques, especially looking at the diagnostic capabilities included on- and off-chip and enabling a fast data analysis.
a) Diagnosis strategies development for quickly feedback failures appearing along Burn In or causing field returns.
b) Design of a SW platform to implement a fast elaboration the failure information wrote in the DB.
Expected target publications: The expected publications are both at Test Conferences like ITC, DATE and ETS and high ranked Journals like Transaction on Computers, Transaction on Emerging Topics in Technologies.
Current funded projects of the proposer related to the proposal: Current funding of the proposer bases on commercial contracts with:
Xilinx (2015-today)
Infineon (2014-today)
Possibly involved industries/companies:The proposer expects to involve the phd students in the research activity carried with Infineon - Xilinx - STMicroelectronics

Title: Accessibility in the Internet of Things
Proposer: Fulvio Corno
Group website: http://elite.polito.it/
Summary of the proposal: The Internet of Things (IoT) is expected to have a significant impact on our daily lives as it will change how we interact with each other, with our objects and our spaces, be they the home or the work place. The IoT provides an opportunity to ensure equal access for everybody: e.g., operating doors and lights through voice makes an IoT-powered environment more accessible to people with physical disabilities and inclusive to many more. Furthermore, web standards for the IoT like the Web of Things extend the open web platform to the physical world and provide a good starting point for making the IoT an enabler, and not a disabler. However, several challenges are still present, ranging from the interoperability of assistive technology with "new" IoT systems to accessibility support in the way we interact with and configure IoT devices, to accessibility guidelines. This PhD proposal aims at investigating these and similar issues and to provide novel and promising strategies, frameworks, tools, and interface that can positively impact the accessibility of contemporary IoT solutions.
Rsearch objectives and methods: The general objective of this PhD proposal is to assess and improve the level of accessibility of new technologies (based on IoT systems). The application areas of such systems range from smart environments (home, workplace), smart transportation (especially public transportation), health and well-being, gaming and entertainment, as well as socialization and communication solutions. Hundreds of new 'connected' products are constantly being proposed by researchers, startups and large industries. While the security and reliability of many of these solutions is a critical issue, this proposal will mainly focus on their interface and usability, in particular for persons with disabilities. New smart devices, joined with novel and intelligent interaction methods, could profoundly empower persons with disabilities in their daily needs. However, many such products are not designed with this need in mind.
One first, minor, research objective is to identify the major accessibility barriers in current IoT systems and devices, both at the level of the adopted interaction algorithms, and of the available user interfaces. In this, we will identify major issues and shortcomings, and propose alternative solutions (at the algorithm level or at the interface level) for improving their accessibility. The research methods will be based on user studies, controlled experiments, and interaction prototypes.
A second, major, research objective is to propose robust and applicable solutions to some of the challenges identified in the previous phase. In this case, we will research, design, prototype and evaluate a novel IoT device or interface able to solve the accessibility barriers for a specific kind of disabilities. Some examples (that will be validated during the study) are: voice assistants for persons with dysarthria or other speech disorders, alternative interfaces to physical IoT devices for persons with motor disabilities, dynamic generation of user interfaces for IoT devices for persons with mild cognitive disabilities (including the elderly), tangible and immersive IoT-based gaming for children with disabilities, etc.
The research methods will combine the techniques adopted in the Human Computer Interaction field (such as Human centered design processes, empirical evaluations, user studies, etc) with know-how about the design and realization of IoT systems (such as distributed systems, embedded devices, mobile interfaces, intelligent algorithms, etc).
Throughout the doctorate, we expect to activate one or two research projects relevant to these topics.
Outline of work plan: The work plan will be organized according to the following four phases.
Phase 1 (months 0-6): literature review about accessibility of digital devices and interfaces; study and knowledge of IoT devices and related communication and programming practices and standards.
Phase 2 (months 6-18): assessment and analysis of accessibility of IoT devices and interfaces. In this phase, some usability assessment and testing, and some accessibility evaluation will be conducted on some significant IoT devices (e.g., conversational assistants, smart thermostats, wearable devices, …) to assess their degree of accessibility for different populations of users. The expected result is the definition of a set of accessibility issues and challenges.
Phase 3 (months 12-24): research, definition and experimentation of novel accessible interfaces for IoT systems. Starting from the results of the previous phase, some of the identified challenges will be addressed in this phase, by proposing new solutions for lowering the accessibility barriers. Such solutions could imply the design of new user interfaces (within the same device or on a different one), or improvement of the intelligent algorithms of the IoT device to compensate for user capabilities.
Phase 4 (months 24-36): the last phase will be twofold. On one hand, the work on addressing the identified challenges will be continued. On the other hand, a robust and large-scale accessible IoT system will be proposed, that includes the proposed technologies and interaction methods, and will be the basis for a comparative assessment of the attained accessibility improvement.
Expected target publications: IEEE Internet of Things Journal, ACM Transactions on Internet of Things, ACM Transactions on Accessible Computing, IEEE Pervasive Computing, IEEE Transactions on Human-Machine Systems, ACM Transactions on Computer-Human Interaction
Current funded projects of the proposer related to the proposal: None at the moment
Possibly involved industries/companies:None at the moment

Title: Recursive set-membership estimation algorithms with application to system identification and adaptive data-driven control design
Proposer: Diego Regruto
Group website:
Summary of the proposal: The research activity of the System Identification and Control (SIC) group has been focused in recent years on the development of convex-relaxation based algorithms for solving classes of polynomial optimization problems specifically arising in the framework of set-membership estimation (set-membership system identification and set-membership data-driven control). Although the effectiveness of the proposed algorithms have been demonstrated by means of a number of real-world applications, their computational complexity prevents the application of such algorithms to problems where the estimation need to be performed in real-time.

The proposed research project is focused on the derivation of recursive version of the set-membership estimation algorithms previously proposed by the research group. Results available in the literature about LP and SDP relaxations for polynomial optimization problem will be carefully analyzed and suitably modified in order to significantly reduce the number of variables and constraints involved in the computation, by accepting a prescribed degree of approximation in the computed solution.

Furthermore, customized interior-point algorithms will be designed in order to speed up the computation as much as possible.

Based on such recursive set-membership estimation algorithms, novel adaptive data-driven control scheme and recursive set-membership identification algorithms will be derived.
Rsearch objectives and methods: The research will be focused on the development of new methods and algorithms for real-time recursive solution to the class of polynomial optimization problems arising in the framework of set-membership estimation theory. The main topics of the research project are summarized in the following three parts.

Part I: Recursive convex relaxations for system identification and control design

This part is focused on the derivation of suitable recursive version of the convex relaxation-based set-membership estimation algorithms previously proposed by the research group. The derived recursive algorithms will be applied to the following problem.

- Recursive set-membership identification of linear models

Set-membership system identification procedures aim at deriving mathematical models of physical systems on the basis of a set of input-output measurements corrupted by bounded noise. In many real-world problems of interest, the parameter of the system to be identified enjoy a time-varying nature. Therefore, identification of such parameter requires application of recursive estimation algorithms. Aim of the research is to applied the derived set-membership recursive estimation algorithms to the problem of identifying linear model from experimental data corrupted by noise on both the input and the output (recursive set-membership errors-in-variables problem).

- Adaptive direct data-driven design of fixed structure controllers:

The aim of this part of the project is to apply the derived recursive set-membership estimation algorithms to the problem of designing a controller directly from a set of input/output data corrupted by noise. Application of the recursive algorithm will implicitly lead to a controller with time-varying parameter, able to adapt in real-time to different experimental situations.

Part II: Development of customized SDP solvers

This part is focused on the reduction of the computational load in solving SDP problems arising from the relaxation of polynomial optimization problems through SOS decomposition. In particular, the peculiar structure of such SDP problems will be exploited in order to design interior-point algorithms more efficient than the ones employed in general purpose SDP, in terms of both memory storage and computational time.

PART III: Application to automotive real-world problems and secure estimation of Cyberphysical systems

The derived algorithms will be applied to a number of real world problems in the automotive field that will arise in the context of a current collaboration together with FCA and CRF. More specifically, the recursive algorithms for estimation and direct adaptive control derived in Part I of the research project will be applied to:

- Design virtual sensors able to provide real-time estimation of variables which are not physically measurable on vehicles, although their real-time knowledge is crucial for improving the performances of control systems acting on the different vehicle subsystems. Example of such quantities includes the vehicle sideslip angle and the clutch position.

- Design of direct data-driven adaptive control algorithms able to force real-time adaptation of the vehicle behavior to different working condition. Examples includes adaptive control of dual-clutch systems subjected to temperature changes and adaptive control of autonomous vehicle (lateral and/or longitudinal control) in presence of different road conditions.

The recursive algorithms for estimation and adaptive control derived in Part I of the project are expected to be general enough to be applicable to a large class of real-world problems. The specific considered problems will be selected in agreement to our industrial partners (FCA/CRF).

The derived algorithms will also be applied to the problem of designing secure real-time state estimation in presence of adversarial attacks in cyberphysical systems. More specifically, the derived recursive set-membership algorithms will be exploited to design observers able to detect in real-time the presence of adversarial attacks corrupting both the amplitude and the time delay of the signals flowing on the networks connecting the sensors and actuators of the cyberphysical systems to the controller and the physical plant respectively. The derived adaptive data-driven control design algorithms will also be exploited for counteracting the effects of the adversarial attacks in such a way to guarantee satisfaction of the control requirements also in presence of the attacks. Different attacker models proposed in the literature will be considered and analyzed.
Outline of work plan: FIRST YEAR

January 1st – June 30th :
the first six months of the project will be devoted to the study of the literature with reference to the subject of recursive estimation, set-membership estimation/identification, direct data-driven control design, convex relaxations of polynomial optimization problems, interior-point algorithms for semidefinite programming.

Milestone 1:
report of the results available in the literature; theoretical formulation of the recursive set-membership estimation problem and analysis of the main difficulties/critical aspect/open problems. Results obtained from this theoretical analysis of the problem will be the subject of a first contribution to be submitted to an international conference.

July 1st – December 31st:
the second part of the first years will be devoted to the problem of deriving recursive version of set-membership estimation algorithms for system identification and direct data-driven control previously derived by the research group.

Milestone 2:
Derivation of original recursive algorithms for set-membership identification and direct data-driven control.
Results obtained in this stage of the project are expected to be the core of a paper to be submitted to an International Journal.

SECOND YEAR

January 1st – June 30th :
the first half of the second year of the project will be focused on the study of the interior-point algorithms available in the literature for the solution of convex optimization problem (with particular focus on semidefinite problems).

July 1st – December 31st:
the objective of this part will be to develop new efficient interior-point algorithms, specifically tailored to the characteristics of the semidefinite programming problems obtained from convex relaxation of the considered optimization problems, arising in the framework of recursive set-membership estimation theory.

Milestone 3:
development of new interior-point algorithms and comparison with the existing general-purpose softwares for semidefinite programming.

THIRD YEAR

January 1st – December 31st:
the last year of the project will be devoted to apply the derived algorithms to modeling, identification and control of a real dual-clutch system and other real-world problems from the automotive field. The problem of secure state estimation in cyberphysical systems will be also considered by exploiting the derived algorithm to the problem of real-time joint states and attacks estimation.
Expected target publications: JOURNALS:
IEEE Transactions on Automatic Control, Automatica, IEEE Transactions on Control System Technologies, System and Control Letters, International Journal of Robust and Nonlinear Control.

CONFERENCES:
IEEE Conference on Decision and Control, American Control Conference, IFAC Symposium on System Identification, IFAC World Congress
Current funded projects of the proposer related to the proposal:
Possibly involved industries/companies:Fiat Chrysler Automobile (FAC) and Fiat Research Center (CRF), which have collaborated with the System Identification and Control (SIC) group continuously in the last decade, will be involved in the application of the derived algorithms to different modeling, identification and control problems arising in the automotive fields.

Title: From data mining to data "storyboarding"
Proposer: Tania Cequitelli
Group website: http://dbdmg.polito.it
Summary of the proposal: In today's world, data value is severely undermined by our inability to translate them into actionable knowledge. Up to now, a lot of research efforts have been devoted to enhancing the effectiveness and efficiency of analytics algorithms. However, their exploitation is still a multi-step process requiring a lot of expertise to be correctly performed. To streamline the knowledge extraction process and enhance the friendliness of data analytics tasks, the PhD student should design and develop a new generation of data analytics solutions to automatically discover descriptive, predictive and prescriptive models hidden in the data without requiring the intervention of the data scientist. A key feature of such solutions should be the automation of the full data-analytic workflow, from heterogeneous data ingestion to the result presentation. They should automatically set the analysis end goal, perform the whole process, and visualize the extracted knowledge in a human-friendly way.
End users, such as domain experts, will be provided with the "story" of their data
• automatically orchestrated by picking the most meaningful results among the plethora of parameters, choices, and trick;
• carefully presented to make the knowledge human-readable and exploitable, so that domain experts can focus on putting such knowledge into actions.
Rsearch objectives and methods: To automatically create the data storyboard in a human-friendly and actionable way, the following research objectives will be addressed:

Data characterization.
To understand structures and models hidden in a data collection, a set of descriptive metrics will be defined by exploiting unconventional statistical indexes and new algorithms to model underlying data structures.

Algorithm selection and optimization.
In the literature several alternative algorithms are available for performing a given data mining task, and in most cases no algorithm is universally superior. To automatically drive the exploration of the search space at different levels (e.g., algorithm class, implementation, parameter setting, etc.) a set of interestingness metrics will be studied to evaluate and compare the meaningfulness of the discovered knowledge.

Knowledge navigation, visualization and exploitation.
The data mining process performed on databases may lead to the discovery of huge amounts of apparent knowledge, that is usually hard to harness. Nevertheless, an in-depth analysis may be required to pick the most actionable and meaningful bits. The characterization of the knowledge significance in terms of innovative, possibly unconventional, transparent criteria will be addressed to rank the results so that the most relevant pieces of information can emerge. Data “stories” based on innovative visualization frameworks will be designed to help end users capture the full data processing flow and foster actionable knowledge exploitation.

Self-learning methodologies based on human-in-the-loop.
To enhance the self-learning capabilities of the proposed solutions, user interactions with the presented data “stories” will be used to collect feedbacks and re-train models for the discovery of interesting end-goals. By providing a way of exploiting user interactions, the overall system can be easily customized and adapted to different application scenarios, while the overall design is kept as general-purpose as possible.
Outline of work plan: During the 1st year, the candidate will study state-of-the-art systems addressing the enhancement of the friendliness of data mining and machine learning algorithms. A set of descriptive metrics will be defined to model data distributions. Since several aspects influence data distributions, the Phd student will draft for each data type (e.g., numerical data, categorical data, short text) an innovative criterion to model data distribution by exploiting unconventional statistical indexes. Furthermore, the candidate will study and define an innovative optimization strategy to properly configure data analytics algorithms.

During the 2nd year the candidate will define innovative metrics to select the most meaningful knowledge that can be effectively transformed into actions by domain experts. A first release of the data storyboard will be defined to provide end users with knowledge presented in a human-friendly way.

During the 3rd year the candidate will study and define a novel algorithm to support a self-learning methodology exploiting a KDB (Knowledge DataBase) able to select the most relevant and meaningful knowledge according to feedbacks from users, both active (human-in-the-loop analytics) and passive (data storyboard navigation patterns).

During the 2nd-3rd year the candidate will assess the proposed solutions in real-life application domains based on the candidate interest, research contracts, funded projects, and available datasets. All these application contexts will be key to disseminate the proposed solutions not only in the primary venues (conferences and journals), but also towards industries and the society, fostering cross-domain fertilization of research results, and new collaborations with other research stakeholders.
Expected target publications: Any of the following journals
IEEE TKDE (Trans. on Knowledge and Data Engineering)
ACM TKDD (Trans. on Knowledge Discovery in Data)
ACM TOIS (Trans. on Information Systems)
ACM TOIT (Trans. on Internet Technology)
Information sciences (Elsevier)
Expert systems with Applications (Elsevier)
Engineering Applications of Artificial Intelligence (Elsevier)
Journal of Big Data (Springer)

IEEE/ACM International Conferences
Current funded projects of the proposer related to the proposal: Progetto di ricerca regionale DISLOMAN (Dynamic Integrated ShopfLoor Operation MANagement for Industry 4.0)
Contratto di ricerca con Enel Foundation (Classification of residential consumers based on hourly-metered consumption data)
Possibly involved industries/companies:None

Title: Urban intelligence
Proposer: Silvia Chiusano
Group website: http://dbdmg.polito.it
Summary of the proposal: Due to the rapid development of cities and the increase in urban population, past decades have seen many urban issues arise, such as traffic congestion, energy shortages and air pollution. At the same time, an increasing volume of a variety of data is generated and collected with the help of new technologies.

Urban intelligence entails the acquisition, integration, and analysis of big and heterogeneous data collections generated by a diversity of sources in urban spaces to profile the different facets and issues of the urban environment.

Digging deep urban data collections can unearth a rich spectrum of knowledge valuable to characterize citizen behaviours, identify weaknesses and strengths of the services provided in the urban area as well as improve the quality of these services or even devise new ones. However, data analytics on these data collections is still a daunting task, because they are generally too big and heterogeneous to be processed through data analysis techniques currently available. Consequently, from a data science perspective, data emerging from today's urban environment give rise to a lot of challenges that constitute a new inter-disciplinary field of research.
Rsearch objectives and methods: The PhD student will work on the study, design and development of proper data models and novel solutions and for the acquisition, integration, storage, management and analysis of big volumes of heterogeneous urban data.

The research activity involves multidisciplinary knowledge and skills including database, machine learning techniques, and advanced programming. Different case studies in urban scenarios such as urban mobility, citizen-centric contexts, and healthy city will be considered to conduct the research activity. The objectives of the research activity consist in identifying the peculiar characteristics and challenges of each considered application domain and devise novel solutions for the management and analysis of urban data for each domain. More urban scenarios will be considered with the aim of exploring the different facets of urban data and evaluating how the proposed solutions perform on different data collections.

More in detail, the following challenges in the context of data analytics will be addressed during the PhD, such as:
- Suitable data fusion techniques and data representation paradigms should be devised to integrate the heterogeneous collected data into a unified representation describing all facets of the targeted domain. For example, urban data are usually acquired by means of different sensor networks deployed in the city. Since these data are often collected with different spatial and temporal granularities, suitable data fusion techniques should be devised to support the data integration phase, and provide a spatio-temporal alignment of collected data.
- Adoption of proper data models. The storage of heterogeneous urban data collections requires the use of alternative data representations to the relational model such as NoSQL databases (e.g., MongoDB).
- Design and development of algorithms for big data analytics. Urban data is usually charaterized by spatio-temporal coordinates describing when and where data has been acquired. Spatio-temporal data has unique properties, consisting of spatial distance, spatial hierarchy, temporal smoothness, period and trend, which entails the design of suitable data analytics methods. Moreover, huge volume of data demands the definition of novel data analytics strategies also exploiting recent analysis paradigms and cloud based platforms as Hadoop and Spark.
- Proper strategies will be also devised for data and knowledge visualization.
Outline of work plan: 1st Year. The PhD student will review the recent literature on urban computing to identify the up-to-date research directions and the most relevant open issues in the urban scenario. Based on the outcome of this preliminary explorative analysis, an application domain, such as urban mobility or urban air pollution, will be selected as a first reference case study. This domain will be investigated to identify the most relevant data analysis perspectives for gaining useful insights and to assess of main data analysis issues. The student will perform an exploratory evaluation of state-of-the-art technologies and methods on the considered domain, and will present a preliminary proposal for the optimization techniques of these approaches.
2nd and 3rd Year. Based on the results of the 1st year activity, the PhD student will design and develop a suitable data analysis framework including innovative analytics solutions to efficiently extract useful knowledge in the considered domain, aimed at overcoming weaknesses of state-of-the-art methods. Moreover, during the 2nd and 3rd year, the student will progressively consider a larger spectrum of application domains in the urban scenario. The student will evalute if and how his/her proposed solutions can be applied to the new considered domains as well as he/she will propose novel analytics solutions.
During the PhD, the student will have the opportunity to cooperate in the development of solutions applied to the research projects on smart cities. The student will also complete his/her background by attending relevant courses. The student will participate to conferences presenting the results of his/her research activity.
Expected target publications: Any of the following journals
IEEE TKDE (Trans. on Knowledge and Data Engineering)
ACM TIST (Trans. on Intelligent Systems and Technology)
Journal of Big Data (Springer)
Expert Systems With Applications (Elsevier)
Information Sciences (Elsevier)

IEEE/ACM International Conferences
Current funded projects of the proposer related to the proposal: CANP (La Casa nel Parco): Regione Piemonte, Piattaforma Tecnologica “Salute e Benessere”
S[m2]ART (Smart Metro Quadro): Bando Smart City and Communities; Ente finanziatore: MIUR (Ministero dell’istruzione, Università e Ricerca)
MIE (Mobilità Intelligente Ecosostenibile): Bando Cluster Tecnologico Nazionale “Tecnologie per le Smart Communities”; Ente Finanziatore: MIUR (Ministero dell’istruzione, Università e Ricerca)
Possibly involved industries/companies:

Title: Visual domain generalization through agnostic representations
Proposer: Barbara Caputo
Group website: https://scholar.google.it/citations?hl=en&user=mHbdIAwAAAAJ
Summary of the proposal: The ability to generalize across visual domains is crucial for the robustness of artificial recognition systems: self-driving cars are requested to identify pedestrians, traffic signs and other moving vehicles regardless of the weather and illumination conditions; robots deployed in domestic environments should be able to recognize kitchen supplies and utensils in different houses, and so forth. Although many training sources may be available in real contexts, the access to even unlabeled target samples cannot be taken for granted. This makes standard unsupervised domain adaptation difficult to use outside of research settings. This thesis will investigate how to exploit multiple sources by hallucinating a deep visual domain composed of images, possible unrealistic, able to maintain categorical knowledge while discarding specific source styles. We will use modern deep learning approaches, rooted in the adversarial learning paradigm, aiming to find principled solutions able to work on a wide range of scenarios, from character recognition to objects categorization, to semantic segmentation. We will test our findings on real-world scenarios, ranging from automatic license plate recognition to human robot interaction, to autonomous driving cars.
Rsearch objectives and methods: Domain Adaptation (DA) is at its core the quest for principled algorithms enabling the generalization of visual recognition methods. Given at least a source domain for training, the goal is to obtain recognition results as good as those achievable on source test data on any other target domain, in principle belonging to a different probability distribution.
Solving this problem will represent a major step towards one of the key goals of computer vision, i.e. having machines able to answer the question ‘what do you see?’ in the wild; hence, its increased popularity in the community over the last years. Since its definition, the most popular instantiation of the problem has assumed to have access to annotated data from a single source domain and to unlabeled data from a different target domain. Still, there has been recently a growing interest on how to leverage over multiple sources for adaptation and for domain generalization (DG) when it is not possible to access target data of any sort a priori.

This thesis will attack the domain generalization problem by hallucinating samples of a latent pixel space, that we call ‘agnostic’, among shared domains. Our intuition is that the notion of visual cross-domain generic information is intuitive yet ambiguous, as ground truth examples of pure semantic images without a characteristic style do not exist. Thus, while it is possible to interpret the produced samples as capturing domain agnostic knowledge, it should be clear that they are built for the network’s benefit only and we do not expect them to be pleasant to the human eye. Preliminary work [a] show that in scenarios with limited visual variability it is possible to learn such agnostic images by letting the network learn what this generic information is through a direct mapping guided by adversarial adaptive constraints. These constraints are applied directly on the agnostic space, rather than on standard images that always contain specific information. This thesis will move forward by exploring deep mapping architectures able to leverage over the power of pre-trained architectures, that have shown to be a key ingredient in the success of deep vision so far, and more realistic domain generalization scenarios, where the visual classes among the different source domains are not fully aligned and/or not all source data available are fully annotated. In order to make the algorithms usable in realistic scenarios, we will strive for solutions scalable with respect to the number of classes, and with deep architectures with a minimal number of parameters.

As applicative domains, we will specifically consider Optical Character Recognition, applied to the automatic reading and recognition of license plates and/traffic signs from wearable seeing devices, surveillance cameras and self-driving cars.

[a] F. M. Carlucci, P. Russo, T. Tommasi, B. Caputo. Agnostic Domain Generalization. arXiv preprint arXiv:1808:01102.
Outline of work plan: The research work plan will be articulated as follows:

M1-M3: Implementation and testing of [a] on reference benchmark databases in character recognition, object categorization and semantic segmentation; implementation and testing of relevant baselines in the literature.

M4-M12: Implementation of deep architectures for the generation of agnostic domains through deep variational autoencoders; development of novel loss function for ensuring convergence and enabling the use of pre-trained convnets. Assessment of work on the established benchmarks. Writing of scientific report on findings of Y1.

M13-M24: Implementation of deep architectures for the generation of agnostic domains scalable with respect to the number of source domains and classes. Assessment of work on the established benchmarks. Writing of scientific report on findings of Y2.

M25-M36: Implementation of final deep architecture for the generation of agnostic domains incorporating the best results from Y1 and Y2, with principled reduction of the internal parameters of the network. Assessment of work on the established benchmarks. Writing of scientific report on overall findings of the project. Writing of PhD thesis.
Expected target publications: It is expected that the scientific results of the project will be reported in the top conference in the field of computer vision (IEEE CVPR, IEEE ICCV, ECCV). At least one journal publication is expected on one of the following international journals: IEEE PAMI, IJCV, CVIU.
Current funded projects of the proposer related to the proposal: ERC RoboExNovo
Possibly involved industries/companies:Italdesign, TIM

Title: Investigation of innovative IoT technologies for complex manufacturing process modelling and optimization
Proposer: Enrico Macii
Group website: http://eda.polito.it
Summary of the proposal: The aim of this research proposal is to investigate innovative solutions aiming to leverage the potential of Internet-of-Things technologies within the manufacturing environment, specially focusing on complex industrial processes. In order to strengthen the link between academic research and industrial applicability, aspects such as readiness, cost, robustness, reliability and flexibility must be taken into consideration during the investigation. The research candidate will explore relevant industrial case studies with identified industrial partners. Taking inspiration from the selected case studies as well as their requirements and objectives, the research will focus on the evaluation of state-of-art Industrial IoT (IIoT) solutions concerning process data collection and analysis. The objective is to evaluate what is missing to develop suitable process models and including the IoT system in the process optimization loop. The core of the research program will then be the design of IoT ecosystems for complex manufacturing process optimization, exploiting learning-based algorithm to improve process parameters. It is of uttermost importance that process models are flexible and reusable. From these aspects, black-box or hybrid models have to be explored. Hybrid models are based on a few parameters, that can be characterized using data collected from the field. Black box models are fully learning based. In this sense, sensor data are used either for model characterization and as input to a parallel virtual process model. Finally, particular attention must be paid to the actual effectiveness of such solutions in increasing the industrial system's competitiveness. Lessons learned from past projects and applications have demonstrated how unless these solutions are developed within a clear continuous improvement framework (such as Lean Production, TPM, …), rather than increasing the competitiveness of the industrial system they contribute to digitalize its losses.
Rsearch objectives and methods: The objectives of such a research project include:

1) Understanding manufacturing process characteristics, including: i) understanding the organization of a manufacturing process (material, equipment and manpower), of its KPI (quality and productivity), of process control; ii) Understanding the needs of the actors of a manufacturing process (operators, maintenance, supervisors, process engineers, management); iii) Identifying case studies on which to carry out the investigation, and justifying such choice.

2) Evaluation of strengths and weaknesses of the current state-of-the-art IoT solutions for manufacturing process data collection and characterization. The evaluation will be carried out focusing on how the current systems can be profitably used to collect and analyse process parameters to create effective data-driven process models.

3) Development of a data-driven manufacturing process modelling framework, able to fastly and flexibly characterize manufacturing processes. The objectives are the evaluation of the proposed solution in term as well as the understanding of the opportunities and threats of developing and implementing such framework.
Outline of work plan: Year 1: Exploration of manufacturing world and IIoT solutions.

- Identification of case studies of manufacturing processes
- Analysis of state-of-art IoT technologies for data collection and process modelling
- Black box and hybrid process models, parameters characterization
- Streaming data processing techniques, including stream data filtering and complex event processing
- First application on identified case studies, identification of opportunities and threats of implementing IoT solutions, and of the conditions which determine these opportunities and threats
- Identification of critical characteristics capable of satisfying these needs

Year 2: Development of a IIoT framework for industrial process characterization.

- Development of IoT-based systems for data collection within the investigated process
- Development of data filtering strategies to enable model characterization.
- Development of techniques for parameter estimation, including learning-based solutions for hybrid or black box model characterization.
- Development of innovative machine learning approaches within the investigated context, capable of optimizing process parameters.
- Greater participation in conferences, conference publications and journal publications (content focusing on developed solutions)

Year 3: Development of parallel virtual models and evaluation.

- Development of virtual models using on-line process inputs running in parallel with the real models.
- Investigation of the clarity and effectiveness of the feedback from the system to the human organization running the industrial process
- Investigation of the effectiveness of the system in guiding the decision-making process and in supporting the verification of the implemented actions
Expected target publications: The results of the research will produce methodologies for manufacturing process modelling algorithms for data filtering and algorithms for parameters exploration and optimization. Such results are expected to be published in journals and conference proceedings about IoT systems and devices, data engineering and Industrial applications.

Possible list of journal and conferences:
- IoT and embedded systems journals (e.g., IEEE IoT Journal, Transactions on Embedded Computing Systems)
- Machine learning and algorithms journals (e.g., IEEE transactions on knowledge and data engineering, expert systems with applications)
- Application-oriented journals (e.g., IEEE transactions on industry applications, transactions on industrial informatics)
Current funded projects of the proposer related to the proposal: DISLOMAN (Regione Piemonte), STAMP (Regione Piemonte), AMABLE (H2020), SERENA (H2020), MANUELA (H2020), R3POWER-UP (ECSEL)
Possibly involved industries/companies:- Automotive companies from the Turin area
- Other companies from the automotive supply chain in the Turin area and in the rest of Italy
- Other companies involved in continuous improvement programs in Italy
- Other industrial SMEs from the Turin area

Title: Domain Generalization via Self-Supervised Learning
Proposer: Tatiana Tommasi
Group website: https://scholar.google.com/citations?user=ykFtI-QAAAAJ&hl=en
Summary of the proposal: Deploying the perception system of an autonomous vehicle in new environments compared to its training setting may lead to dramatic failures owing to the shift in data distribution. This is just one example of the generalization issues that affects also strong artificial learners as deep neural networks.
The ability to generalize is a hallmark of human intelligence related to the skill of merging knowledge from both supervised and unsupervised learning. This is particularly effective because supervised learning can never be exhaustive and thus learning autonomously allows to discover invariances and regularities that help to generalize.
This thesis will investigate how to formalize this skill for artificial agents by training a model that jointly learns semantic labels in a supervised fashion and broadens its understanding of the data by solving tasks from self-supervised signals. We will start by exploiting spatial co-location of patches: an intrinsic visual information that can be used to solve jigsaw puzzles on images from multiple domain sources.
We will use modern deep learning approaches defining tailored multi-task learning strategies with the aim of defining principled solutions able to work on a wide range of real world scenarios and possibly exploiting heterogeneous data sources from images to text and motor information in robotics applications.
Rsearch objectives and methods: The goal of Domain Generalization (DG) is that of defining learning methods that can perform uniformly well across multiple data distributions. The standard setting supposes access to several data sources with specific biases (e.g. different datasets, visual styles) and the main challenge is being able to distill transferrable general knowledge for future applications on new unseen domains. Recent DG approaches propose either to decompose data information into domain specific and domain agnostic, then discarding the first one, or to exploit complex meta-learning solutions that can be integrated into simple architectures but are hard to scale to better performing networks (e.g. ResNet).
This thesis proposes to tackle domain generalization with a new strategy based on self-supervised learning. Indeed several works have shown that visual invariances and regularities can be captured by image coloring, artifact discovering, video-frame ordering and spatial co-location of patches. All these tasks exploit intrinsic data signals that do not need human supervision and the learned deep features can be effectively transferred to new related tasks. Preliminary results show that reordering image patches as in a jigsaw puzzle and jointly training supervised classification models allow to regularize the learning process over multiple domains with significant generalization effects on object recognition.
This thesis will further investigate over these initial results and move forward in different directions. First of all we will focus on the jigsaw puzzle problem and explore different architectural solutions to define this task (patches recomposition at input/intermediate/output level; objective loss classification/ranking; size and shape of the patches and their relation to object parts) and to integrate it with the supervised classification task (multi-task networks with different task connection points; various data augmentation procedures). As a second step we will further improve generalization by adding tailored adaptive losses and adversarial conditions also introducing principled constraints on the number of learning parameters. As a third step we plan to extend the study to other self-supervised tasks besides jigsaw puzzle (e.g. frame ordering, artifact discovery and inpainting) and to investigate their applicability when dealing with heterogeneous sources (e.g. RGB and Depth images, Images and grasping motor information, Images and textual attributes or captions).

As applicative domains we will tackle object classification and detection, person re-identification and action recognition from data acquired by robotics agents and surveillance cameras. For the heterogeneous extensions we will consider visual question answering for human-robot interaction and robotic object manipulation.
Outline of work plan: The research work plan will be articulated as follows:

M1-M3: Implement deep learning model for jigsaw puzzle from existing literature and study patch recomposition solutions at different network points (input image, intermediate convolutional layers, fully connected output). Implement and test relevant baselines. Preparing a robotic domain generalization testbed from images collected with iCub platform under different translation, scale and 2D/3D rotations conditions.

M4-M12: Implement a multi-task architecture to jointly learn object classification and jigsaw puzzle. Optimize the jigsaw puzzle loss considering different task formalizations (classification/ranking) and the combination of multiple patch sizes. Validate different task splitting points and data augmentation strategies. Assessment of work on the established benchmarks. Writing of scientific report on findings of Y1.

M13-M24: Extend the obtained implementation by integrating adaptive losses and adversarial conditions. This should be done in a principled way to ensure network convergence and to avoid an extreme growing of the architecture depth or of the number of dedicated network branches. Assessment of work on the established benchmarks. Writing of scientific report on findings of Y2.

M25-M36: Extend the learning model to other self-supervised task. Analyze the domain generalization effect when moving from spatial to temporal image co-location by studying frame ordering, and when integrating generative solutions for image inpainting. Check the applicability of the proposed solution when scaling the number and nature (visual, textual, motor signals) of sources. Assessment of work on the established benchmarks. Writing of scientific report on overall findings of the project. Writing of PhD thesis.
Expected target publications: It is expected that the scientific results of the project will be reported in the top conferences in the field of computer vision (IEEE CVPR, IEEE ICCV, ECCV). At least one journal publication is expected on one of the following international journals: IEEE PAMI, IJCV, CVIU.
Current funded projects of the proposer related to the proposal: CHIST-ERA project BURG: Benchmarks for UndeRstanding Grasping (2019-2021)
Possibly involved industries/companies:

Title: First person action recognition from multi-modal data
Proposer: Barbara Caputo
Group website: https://scholar.google.it/citations?hl=en&user=mHbdIAwAAAAJ
Summary of the proposal: While the vast majority of existing digital visual data has been acquired from a third person perspective, egocentric vision is soon going to become a key technology for supporting new developments in assistive driving, enhanced cognition and many more applications. This in turn will require the development of visual recognition algorithms able to deal with the challenges of this new scenario, from transfer learning across different actors to action anticipation, and so forth. This PhD thesis will explore first person action recognition on RGB-D data, i.e. when not only standard images are available, but also depth information. The thesis will develop new algorithms for action recognition that are robust to different visual domains, so to be used on various applications, from industrial robot applications to assistive driving.
Rsearch objectives and methods: Automated analysis of videos for content understanding is one of the most challenging and well researched areas in computer vision and multimedia and possesses a vast array of applications ranging from surveillance, behavior understanding, video indexing and retrieval, human-machine interaction, etc. The majority of researchers working on video understanding problem concentrates on action recognition from distant or third person views while egocentric activity analysis has been investigated more recently. Action recognition involves identifying a generalized motion pattern of hands such as take, put, stir, pour, etc. whereas activity recognition concerns more fine-grained composite patterns such as take bread, take water, put sugar in coffee, put bread in plate, etc. For developing a system capable of recognizing activities, it is pertinent to identify both the hand motion patterns as well as the objects on to which a manipulation is being applied to. Although commonly forgotten, depth information should be taken into consideration, both for better understanding actions and their context, and also for making it possible to translate results in this field into human-robot interaction scenarios in a straightforward manner. Moreover, generality across different visual domains is completely ignored in the field at the moment, in spite of being one of the key challenges today to fully bring computer vision into commercial products. This PhD thesis will tackle these problems, extending previous state of the art first person action recognition architecture to deal also with depth data and to deal with domain adaptation and generalization problems. Progress will be assessed on existing public benchmarks, as well as data collected by the PhD student using a Pupil wearable helmet, in driving and assistive robotic scenarios.
Outline of work plan: The research work plan will be articulated as follows:

M1-M3: Implementation and testing of [a] on reference benchmark databases in first person action recognition; implementation and testing of relevant baselines in the literature.

M4-M12: Implementation of deep architectures, extending [a] by leveraging on results and ideas from [b,c], for first person action recognition from RGB-D data. Testing on public reference benchmarks and comparison against relevant baselines. Writing of scientific report on findings of Y1.

M13-M24: Extensions of the deep architecture obtained in Y1 to domain adaptation and generalization problems. Assessment of work on the established benchmarks. Planning of data collection in assistive driving and human-robot collaboration scenarios. Writing of scientific report on findings of Y2.

M25-M28: Collection of two new databases for RGB-D first person action recognition in assistive driving and human-robot collaboration scenarios. Running of established baselines, as well as of algorithms derived in Y1 and Y2. Writing on scientific report.

M28- M36. Implementation of final deep architecture incorporating the best results from Y1 and Y2. Assessment of work on the established benchmarks and on the new databases created in the project. Writing of scientific report on overall findings of the project. Writing of PhD thesis.

[a] S. Sudhakaran, O. Lanz. Attention is all we need: nailing down object-centric attention for egocentric activity recognition. Proc. BMVC 2018.
[b] F. M. Carlucci, P. Russo, B. Caputo. DE^2CO: Deep Depth Colorization. IEEE Robotics and Automation Letters, 3 (3), 2386-2393.
[c] M. R. Loghmani, M. Planamente, B. Caputo, M. Vincze. Recurrent convolutional fusion for RGB-D object recognition. arXiv preprint, arXiv: 1806.01673
Expected target publications: It is expected that the scientific results of the project will be reported in the top conferences in the field of computer vision (IEEE CVPR, IEEE ICCV, ECCV). At least one journal publication is expected on one of the following international journals: IEEE PAMI, IJCV, CVIU.
Current funded projects of the proposer related to the proposal: RoboExNovo
Possibly involved industries/companies:ItalDesign; IBM; Comau