Research Projects Proposals 2024 (XXXVL Cycle)
Systems Engineering
Title: Distributed and robust control techniques for system of systems: application to vehicle platoons, machine to machine interactions, machine to infrastructure interactions
Proposer: Roberto Sacile
Curriculum: Systems Engineering
Abstract:
In this PhD project, new distributed and robust control techniques will be investigated from a theoretical viewpoint. The domain of application is related to wheeled mobile robots and their control in a context related to the interactions with other vehicles, including aerial ones, and the infrastructure. Theoretical problems related to the identification of the position in indoor environments will also be investigated.
Title: Optimization and control of sustainable districts and active prosumers.
Proposer: Michela Robba
Curriculum: Systems Engineering
Research area(s): Optimization, optimal control, model predictive control, energy communities, sustainable districts, aggregator, energy market.
Abstract:
Significant effort has been expended of late, across the globe, to reduce greenhouse gas emissions and to achieve sustainability which in turn has led to a fundamental change in the management of resources and cities. In the case of the energy sector, the increase in the use of renewable resources and distributed generation, as well as the growing presence of electric vehicles, prosumers, microgrids, and other small-scale distributed resources, have enabled and accelerated restructuring of the electrical grid. In order to carry out optimal management of these distributed resources and address associated control problems, new models, methods, and technologies are necessary [1]. Concomitantly, new legislation, actors, and market mechanisms that support such an emerging grid structure are needed [2]. A novel concept that has been recently introduced in the context of smart energy management in cities is that of an Energy Community (EC), i.e., a set of residential or small commercial agents, each acting as prosumer and generally including generation (electric and thermal), storage units such as batteries, and flexible loads. Another new market entity is the aggregator, i.e. a market actor in charge of interacting with the Transmission System Operator (TSO) to reduce a load of a portion of territory through the coordination of different prosumers and users.
The proposed PhD research activity will fall within this framework. In particular, the following main objectives/activities can be listed:
Link to personal homepage
http://www.dibris.unige.it/robba-michela
References:
[1] M. Garcia, H. Nagarajan, and R. Baldick, “Generalized convex hull pricing for the ac optimal power flow problem,” IEEE Transactions on Control of Network Systems, vol. 7, no. 3, pp. 1500–1510, 2020.
[2] G Ferro, M Robba, R Haider, AM Annaswamy, “A distributed optimization based architecture for management of interconnected energy hubs,” IEEE Transactions on Control of Network Systems, 2022.
Supervisor(s): Silvia Siri
Title: Optimization algorithms for smart logistics systems based on mobile robots with innovative electric charging methods
Keywords: Optimization, Scheduling, Smart Charging, Automated Warehouses, Freight Logistics
Curriculum: Systems Engineering
Abstract:
Researchers have been developing new solutions for logistics systems for some time, with applications for automated warehouses, city logistics, intermodal freight movements. This project will focus on smart logistics systems based on the use of mobile robots that have to deliver products in internal and external environments, provided with innovative charging systems integrated with renewable power plants and electric storage systems. For such systems optimization algorithms are needed in order to integrate scheduling and routing problems (for optimally planning the different tasks and the paths to follow) with smart charging strategies (to optimally exploit the local energy production from renewable sources and the use of stationary storage batteries).
The testing phase of such activities will be developed in the new laboratory to be realized in Savona campus, including mobile robots provided with innovative charging systems, renewable power plants, and electric storage systems within a sustainable building.
External collaborations:
Both with Academia and local Industries/Institutions.
Supervisor(s): Cecilia Pasquale, Simona Sacone, Silvia Siri
Title: Control algorithms for smart and sustainable mobility systems
Keywords: Connected and Automated vehicles, Electric Mobility, Sustainability, Smart Cities
Curriculum: Systems Engineering
Abstract:
Mobility systems are key components for the design of smart cities [1] and, in general, for improving the efficiency and environmental sustainability of traffic networks [2]. This is achievable thanks to innovative solutions that exploit technologies to inform transport users, to analyze and monitor the system’s behavior, to optimize routes and travel choices, to control all the transport systems for both passengers and freight. Smart mobility systems must be designed with the aim to follow the ambitious goals fixed by many institutions worldwide for environment protection and sustainable development [3].
In this project, control algorithms will be studied and developed for smart mobility systems that exploit innovative solutions such as connected and automated vehicles, truck platooning [4], electric transport systems [5], mobility-as-a-service systems, smart delivery systems for freight.
References
[1] Q.-S. Jia, H. Panetto, M. Macchi, S. Siri, G. Weichhart, Z. Xu (2022), “Control for smart systems: Challenges and trends in smart cities”, Annual Reviews in Control, 53, pp. 358-369, 2022
[2] G. Lyons, “Getting smart about urban mobility - Aligning the paradigms of smart and sustainable”, Transportation Research Part A, 115, pp. 4-14, 2018
[3] European Commission, “EU's Strategic Transport Research and Innovation Agenda (STRIA)”, hQps://ec.europa.eu/info/research-and-innova-on/research-area/transport/stria\_en
[4] A. Bozzi, S. Graffione, C. Pasquale, R. Sacile, S. Sacone, S. Siri, “A hierarchical control scheme to improve the travel performance of truck platoons in freeways”, ITSC Conference, pp. 2063-2068, 2022
[5] C. Pasquale, S. Sacone, S. Siri, A. Ferrara, “Traffic-prediction-based optimal control of electric and autonomous buses”, IEEE Control Systems Letters, 6, pp. 3331-3336, 2022
External collaborations:
Both with Academia and local Industries/Institutions.
Massimo Paolucci | Smart and green scheduling approaches for manufacturing industry | Green scheduling, multi-objective optimization, metaheuristic algorithms, matheuristic algorithms | Ingegneria dei sistemi |
In recent years energy-efficient scheduling has become an important topic in the scientific literature, as it has taken a key role in ensuring sustainability of manufacturing industry through the reduction of energy consumption and carbon emissions. Rethinking the production processes under a sustainable lens, and simultaneously fostering environment-aware consumption practices in customers, appear to be more and more necessary. To this end, one of the first actions undertaken by energy suppliers consisted of flattening the peaks of demand in power plants by means of strategies aimed at reducing the high economic burdens related to the generation of high energy loads in short periods of time and, consequently, the environmental impact related to energy production. A possible strategy consists of the Time-of-Use (TOU) pricing policy, that spur electricity usage at off-peak hours by means of low prices, while penalizing peak hours with higher prices. In manufacturing, TOU-based energy tariffs can be taken into account by carefully rescheduling the production processes during periods characterized by low energy supply costs.
In such a context, the proposed research aims at developing both exact and heuristic approaches for scheduling problems that require the simultaneous optimization of multiple objectives, in particular including the minimization of the energy cost/consumption and the optimization of performance indexes related to the effectiveness of production and customer satisfaction (e.g., the makespan or tardiness minimization). This research should focus on the class of parallel machine scheduling, progressively including qualifying features to be able to model actual industrial requirements. The heuristic approaches to be designed and experimented can consist in both matheuristic, in particular based on mixed integer programming formulations, and metaheuristic algorithms.
References:
Anghinolfi, D., Paolucci, M., Ronco, R. 2020. A bi-objective heuristic approach for green identical parallel machine scheduling, European Journal of Operational Research, Elsevier, vol. 289(2), 416-434, DOI:10.1016/j.ejor.2020.07.020.
Catanzaro, R. Pesenti, R. Ronco. 2023. Job Scheduling under Time-of-Use Energy Tariffs for Sustainable Manufacturing: A Survey. European Journal of Operational Research, vol. 308(3), 1091-1109, DOI:10.1016/j.ejor.2023.01.029.
Massimo Paolucci | Optimal management of electric vehicles for green logistics | Electric vehicle routing, multi-objective optimization, metaheuristic algorithms, matheuristic algorithms | Ingegneria dei sistemi |
Due to the recent concerns on the climate changes, attention is being paid to the road transport, responsible for about 70% of the harmful emissions of the transportation sector. To this purpose, several countries are promoting green vehicles (e.g., electric vehicles (EVs)). However, especially in the mid-haul logistics, where the driving range of medium-duty EVs cannot usually cover the distances, EVs may require recharging during their trip. Considering that the recharging stations (RSs) are not currently widespread on the territory, this requires properly planning their routes, including the stops at RSs. Such a decision problem is referred in the literature as Electric Vehicle Routing Problem with Time Windows, E-VRPTW. It aims at routing a fleet of EVs, to serve all customers within their time windows, with possible recharges en-route, minimizing the total travelled distance. EVs start from and return to a common depot or different depots, and their trips have to be completed within a maximum time. In order to generate feasible routing plans, realistic energy consumption models are needed, as for example the one assuming that the energy consumption depends on EV speed and load. Moreover, the EV speed may be also a decision variable, varying in a given range, in particular when it can suggested to drivers thanks to a driver-assistance system (ADAS) or for future logistics based on autonomous vehicles. The objective of planning is to minimize the fixed costs of using EVs, the energy costs and the drivers’ wage costs. This research project aims at designing innovate models and solution methods to effectively manage freight logistic services based on fleets of EVs. Alternative models based on Mixed Integer Linear Programming (MILP) will be studied, and new decision algorithms that exploit in intelligent ways such models (both math-heuristics as Kernel Search or meta-heuristics as Genetic Algorithms) in order to generate high quality solutions in an acceptable computation time.
References:
Bruglieri, M., Paolucci, M., Pisacane, O. 2023. A matheuristic for the electric vehicle routing problem with time windows and a realistic energy consumption model. Computers Operations Research, 157 , 106261. DOI:10.1016/j.cor.2023.106261.
Ferro, G., Paolucci, M., Robba, M. 2020. Optimal charging and routing of electric vehicles with power constraints and time-of-use energy prices. IEEE Transactions on Vehicular Technology, 69, 14436–14447. DOI:10.1109/TVT.2020.3038049.
Computer Science
Supervisor(s): Matteo Moro, Nicoletta Noceti, Francesca Odone
TItle: Advancing Learning Using Privileged Information (LUPI)
Keywords: Computer Vision, Machine Learning, Medical Image Analysis
Curriculum: Computer Science
This PhD research proposal focuses on advancing the Learning Using Privileged Information (LUPI) paradigm [1]. LUPI enhances machine learning models by incorporating additional information during training, which is not available during test, simulating the human learning process where detailed explanations are provided during study but not during exams [2]. This research aims to develop novel deep learning algorithms that leverage LUPI to improve classification accuracy and robustness, particularly in complex medical imaging tasks.
The primary technical objective is to design and implement deep learning architectures that effectively utilize privileged information to enhance feature extraction, model generalization, and classification performance. This involves integrating techniques such as CNNs, attention mechanisms, and transfer learning within the LUPI framework. A key aspect of this research will be the development of efficient training strategies that optimize the use of privileged information.
As a practical example, the proposed methodology will target classification of multiple sclerosis (MS) lesions from magnetic resonance imaging (MRI) [3]. Standard MRI sequences, such as T1 and FLAIR, will be used for testing. Meanwhile, additional quantitative data, including T1 relaxometry and Quantitative Susceptibility Mapping—more informative but also more expensive and challenging to acquire—will serve as privileged information during training. This application will demonstrate the potential of LUPI to enhance diagnostic accuracy and clinical decision-making in medical imaging. By advancing the LUPI paradigm, this research aims to contribute significantly to the field of machine learning and its application in healthcare, providing a foundation for more intelligent and adaptive diagnostic systems.
[1] Vapnik, V., & Vashist, A. (2009). A new learning paradigm: Learning using privileged information. Neural networks, 22(5-6), 544-557.
[2] Garcia, N. C., Morerio, P., & Murino, V. (2018). Modality distillation with multiple stream networks for action recognition. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 103-118).
[3] Elliott, C., Rudko, D. A., Arnold, D. L., Fetco, D., Elkady, A. M., Araujo, D., ... & Fisher, E. (2023). Lesion-level correspondence and longitudinal properties of paramagnetic rim and slowly expanding lesions in multiple sclerosis. Multiple Sclerosis Journal, 29(6), 680-690.
Supervisor(s): Nicoletta Noceti, Francesca Odone
TItle: Understanding the dynamics of social interaction
Keywords: Computer Vision, Machine Learning, Human-Robot Interaction
Curriculum: Computer Science
This research proposal aims to investigate the dynamics of social interaction. The primary objective is to develop a comprehensive framework that leverages machine learning and computer vision to identify, classify, and interpret social interactions from videos in diverse settings.
The study will focus on three key aspects: detecting social cues [1], analysing interaction patterns [2], and evaluating contextual influences on these interactions. Depending on the setting, different video features such as facial expressions, poses, and movements will be employed. If available, additional information will be incorporated into the models using multi-modal approaches [3][4], including audio or inertial data. The integration of contextual information, such as the environment or the cultural background, will also be explored. Among the available families of methodologies, we will explore the use of graphs neural networks and variants, among others.
The research will begin with a comprehensive analysis of the state-of-the-art to gain a thorough understanding of existing approaches and identify potential research directions. This will include a detailed analysis of available datasets (see for instance [5])and the design of acquisition protocols where necessary. The main goal for the remainder of the study will be the design, development, and validation of robust algorithms for social interaction analysis. For validation, various use-cases will be identified and examined, including applications in the robotics domain.
[1] Figari Tomenotti, Noceti, Odone “Head Pose estimation with uncertainty and an application to dyadic interaction detection” CVIU 2024
[2] Chang, F., Zeng, J., Liu, Q., Shan, S., 2023. Gaze Pattern Recognition in Dyadic Communication. In: Proceedings of the 2023 Symposium on Eye Tracking Research and Applications. pp. 1–7.
[3] Trabelsi R., Varadarajan J., Pei Y., Zhang L., Jabri I., Bouallegue A., Moulin P.
Robust multi-modal cues for dyadic human interaction recognition
Proceedings of the Workshop on Multimodal Understanding of Social, Affective and Subjective Attributes, ACM (2017), pp. 47-53
[4] Lee et al “Modeling Multimodal Social Interactions: New Challenges and Baselines with Densely Aligned Representations” CVPR 2024
[5] JRDB-Social: A Multifaceted Robotic Dataset for Understanding of Context and Dynamics of Human Interactions Within Social Groups CVPR 2024
Supervisor(s): Massimiliano Pontil (IIT)
TItle: Machine learning and dynamical systems
Keywords: Machine Learning, Dynamical Systems
Curriculum: Computer Science
Abstract
Dynamical systems are mathematical models used to describe the evolution of state variables over time. These models, often represented by nonlinear differential equations (ordinary or partial), and possibly stochastic, have broad applications in science and engineering. This project will focus on investigating data-driven methods, including those based on reproducing kernel Hilbert spaces and deep neural networks, for analyzing dynamical systems. These methods aim to be efficient and facilitate principled analysis. The research will encompass both theoretical and applied aspects, employing a range of mathematical techniques such as mathematical statistics, stochastic processes, empirical processes, numerical analysis, optimization, and optimal transport. Furthermore, the project will explore potential applications in computational chemistry, particularly within the context of molecular dynamics.
Annalisa Barla | Advanced machine learning methods for the understanding and visualization of complexity in science | machine learning, data visualization, science of science | Informatica |
Complexity is evident in all aspects of our lives, from individual relationships to global phenomena such as pandemics or climate change, and it is clear-both at the individual and societal levels-that understanding and managing complexity is the challenge of the present and the future. Moreover, the multidisciplinary nature of wicked problems (Buchanan, 1992) requires the adoption of new tools that can collect and systematize the contributions of various domain experts, to foster interdisciplinary research (Wilson and Zamberlan, 2015).
Only in this way can we envision the possibility of providing scientists, citizens and stakeholders with the tools they need to make the positive contribution our society requires. In the context of scientific research, complexity pushes researchers out of their comfort zone and to seek the collaboration and expertise of scholars from other fields. This multidisciplinary approach fosters greater innovation and a broader view of the scientific questions being addressed.
A full understanding of the quality of cross-disciplinary collaborations is, therefore, more and more necessary and can be achieved through quantitative metrics and methods.
Collaboration is often analyzed through co-authorship in scientific publications, as shown by several studies (Fagan, 2018). Nevertheless, recent advances in Heterogeneous Information Network (HIN) analysis allow us to consider Academic Collaboration Networks (ACN) as complex systems involving several types of entities (e.g. authors, papers, venues and fields of study) and several types of relationships among them (e.g. co-authorship, author-writes-paper, paper-accepted by-venue and so forth). Being able to represent a collaboration graph in such a way that semantics complexity is preserved may allow to perform tasks and consequently define usable systems able to make the academic community even more connected. More specifically, heterogeneity in collaboration graphs may be leveraged to represent research groups (or, more in general, universities) from different perspectives and this would allow for recommending people, works or research topics characterizing the group to external users (e.g. students looking for scholarships, authors looking for collaboration on specific topics, companies interested in R&D collaborations).
In this research work the candidate will exploit machine learning(Hastie et al 2009), deep learning (Goodfellow et al., 2016), network analytics (Barabási, 2013), Natural Language Processing and data visualization (Schwabish, 2021) to make sense of large-scale scientific data for enhancing scientific communication towards the general public as well as the scientific community
References
- Barabási, A. L. (2013). Network science. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 371(1987), 20120375.
- Buchanan, R. (1992). Wicked problems in design thinking. Design issues, 8(2), 5-21.
- Fagan, J., Eddens, K. S., Dolly, J., Vanderford, N. L., Weiss, H., & Levens, J. S. (2018). Assessing research collaboration through co-authorship network analysis. The journal of research administration, 49(1), 76.
- Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press.
Hastie, T., Tibshirani, R., Friedman, J. H., & Friedman, J. H. (2009). The elements of statistical learning: data mining, inference, and prediction (Vol. 2, pp. 1-758). New York: springer.
Wilson, S., & Zamberlan, L. (2015). Design for an unknown future: Amplified roles for collaboration, new design knowledge, and creativity. Design Issues, 31(2), 3-15.
Schwabish, J. (2021). Better data visualizations: A guide for scholars, researchers, and wonks. Columbia University Press.
Stefano Rovetta | Application of artificial intelligence in industrial manufacturing | Macxhine learning and Computational intelligence | Informatica |
In the manufacturing industry, the ever-increasing availability of data and signals from cameras and sensors allows the application of Artificial Intelligence techniques in an ever-increasing number of lines and levels of operations, thus improving efficiency and product quality, employee safety and also allowing the implementation of virtuous circular economy flows. Among the main ones of AI in manufacturing we can mention: (a) Quality control, where AI algorithms are used for the detection of surface and/or functional defects of industrial products, in order to notify production units of potential production defects that can lead to product quality problems [4,[5]; (b) Predictive maintenance of industrial equipment, where AI algorithms allow to accurately predict the malfunction of the assets [3]; (c) Process monitoring where AI algorithms use signals from sensors and cameras and integrate them with the company database; (d) Sorting of end-of-life product components, in order to optimize the recovery of secondary raw materials [2], [1]. After defining some relevant use cases, publicly available data sets and data sets owned by partner companies will be identified and the most suitable machine learning algorithms to solve the problems addressed will be developed and tested.
References:
[1] Cabri, A., Masulli, F., Rovetta, S., Mohsin, M., “Recovering Critical Raw Materials from WEEE using Artificial Intelligence”, The 21st International Conference on Modelling and Applied Simulation (MAS), 19th International Multidisciplinary Modeling & Simulation Multiconference, pp. 1–5, doi: 10.46354/i3m.2022.mas.023, 2023.
[2] Calaiaro, J. "AI Takes a Dumpster Dive: Computer-vision systems sort your recyclables at superhuman speed," in IEEE Spectrum, vol. 59, no. 7, pp. 22-27, July 2022, doi: 10.1109/MSPEC.2022.9819884.
[3] Ran, Y., Zhou, X., Lin, P., Wen, Y. and Deng, R.A “Survey of Predictive Maintenance: Systems, Purposes and Approaches”. IEEE Communications Surveys & Tutorials, 20, 1-36. 2019
https://arxiv.org/pdf/1912.07383.pdf
[4] Saberironaghi, A., Ren, J., El-Gindy. M., “Defect Detection Methods for Industrial Products Using Deep Learning Techniques: A Review”. Algorithms. 2023; 16(2):95. https://doi.org/10.3390/a16020095
[5] Schlüter, M., et al. “ AI-enhanced Identification, Inspection and Sorting for Reverse Logistics in Remanufacturing,” Procedia CIRP, Volume 98, 2021, Pages 300-305, ISSN 2212-8271
Manuela Chessa (Unige), Danilo Pani (UniCa) | Interaction techniques and XR enabling technologies for special user needs | Human-computer interaction, Virtual Reality, Extended Reality | Informatica |
Nowadays, serious games are widely used for training and assessing people with cognitive and physical disabilities. Nevertheless, some diseases need specific solutions in terms of the design and implementation of interaction and visualization techniques.
This PhD research theme aims to explore and devise innovative solutions to allow robust interaction in multimedia systems, with specific attention to people with special needs.
The research activities will consider interaction modalities based on the combination of computer vision techniques and low-cost and light wearable sensors and the combination of inputs from the upper limbs and the users’ gaze and/or head direction.
The developed multimedia systems will explore immersive and non-immersive virtual reality, and, more in general, extended realty, i.e., the combination of virtual and physical objects to exploit passive haptics.
Moreover, the developed systems would be able to adapt to the users needs, through the online modulation and the specific customization of both contents, visualization and interaction techniques.
Manuela Chessa (Unige), Fabio Solari (Unige) | Interactive environments in extended reality (XR) | Human-computer interaction, Virtual Reality, Extended Reality | Informatica |
Extended Reality (XR) and Mixed Reality (MR) are the new frontiers of human-computer interaction (HCI), combining the potentiality of immersive Virtual Reality (VR) with the real physical world. Several fields of application may benefit from such systems, from industrial contexts, for training and maintenance, to medical ones, for rehabilitation and daycare. Also, entertainment, e.g., videogames or museum applications, is an area where XR technologies are becoming prominent. Technology is now advanced enough to provide us with many devices to visualize XR worlds and track the users. Nevertheless, available systems are still preliminary from the computational point of view. This PhD research theme aims to grow a new researcher able to develop and combine algorithms from Computer Vision, to build a dynamic 3D representation of the real world, with HCI and VR techniques. The final goal should be a coherent XR environment where a user should be able to interact with both real and virtual elements. The user in the XR should show natural, i.e., similar to the corresponding real situations, cognitive and physical behaviors and super-natural experiences must be allowed to overcome the limits of the real world.
References:
Ballestin, G., Chessa, M., & Solari, F. (2021). A registration framework for the comparison of video and optical see-through devices in interactive augmented reality. IEEE Access, 9, 64828-64843.
Chessa, M., & Solari, F. (2021). The sense of being there during online classes: analysis of usability and presence in web-conferencing systems and virtual reality social platforms. Behaviour & Information Technology, 40(12), 1237-1249.
Viola, E., Solari, F., & Chessa, M. (2021). Self Representation and Interaction in Immersive Virtual Reality. In VISIGRAPP (2: HUCAPP) (pp. 237-244).
Valentini, I., Ballestin, G., Bassano, C., Solari, F., & Chessa, M. (2020). Improving obstacle awareness to enhance interaction in virtual reality. In 2020 IEEE Conference on Virtual Reality and 3D User Interfaces (VR) (pp. 44-52). IEEE.
Daniele D'Agostino | Virtual research environments for searching and analysing astrophysical data | Distributed Systems, Artificial Intelligence | Informatica |
Modern telescopes can yield unique insights into the universe. But a huge amount of information remains stored and largely un exploited in the data archives. From an ICT point of view the key issues are represented by the fact that these archives have been developed with rather old technologies, have a limited implementation of the FAIR principles and they lack an effective integration with computational infrastructures and analysis tools.
Starting from the experience gained in previous European projects, the goal of the thesis is two-fold. At first an existing archive will be redesigned to provide a state-of-the-art support to scientists in finding, analysing and post-processing the available data.
Of particular interest the development of (i) innovative tools to allow users to perform customized analysis of archival data, also using machine learning approaches; (ii) process flows, also based on machine learning, to perform complex analysis and cleaning, changing in an adaptive and incremental way as new data are generated and ingested in the archive.
Then the integration of National and European research infrastructures, like the ones provided by Fenix, EGI and the national centers funded by PNRR.
References:
D’Agostino, Daniele, et al. "A citizen science exploration of the X-ray transient sky using the EXTraS science gateway." Future Generation Computer Systems 111 (2020): 806-818.
Kovacevic, Milos, et al. "Exploring X-ray variability with unsupervised machine learning-I. Self-organizing maps applied to XMM-Newton data." Astronomy & Astrophysics 659 (2022): A66.
Francesca Odone, Nicoletta Noceti, Vittorio Murino | Learning multimodal representations and applications | Image Processing, Computer Vision, Machine Learning, Deep Learning | Informatica |
The availability of multimodal data in machine learning presents both opportunities and challenges, and many problems well studied in unimodal settings are still largely unexplored.
This project aims to study, design and develop methods for the efficient representation of multimodal data, especially images/videos, audio and text. We will investigate the use of unsupervised, self-supervised and semi-supervised learning for knowledge transfer, and domain adaptation and generalization, with a particular interest in scenarios with scarce resources, i.e., when annotations may be limited, few/zero-shot, imbalanced or biased data. Environmental (climate, energy) and robotics domains will be a testbed for the application of the devised methodologies to real-world challenges.
Viviana Mascardi | The Intersection of Software Agents, Virtual Reality and Natural Language Processing | Artificial Intelligence, Declarative Technologies, Multiagent Systems, Beliefs-Desires-Intentions, Natural Language Processing | Informatica |
Title: The Intersection of Software Agents, Virtual Reality and Natural Language Processing
Proposer: Viviana Mascardi
Research area: Data Science and Engineering
Curriculum: Computer Science
Description: A large share of modern artificial intelligence focuses on designing and implementing intelligent software agents, namely software artifacts situated in physical or virtual environments, creating agent societies, and showing believable, human-like skills and behaviors. The environment where agents live may be fully virtual -- agents live inside a Virtual Reality and the agent society consists of software agents only; interaction takes place via some suitable agent communication languages --, real -- agents perceive and modify the real environment via sensors and effectors and interact with humans using natural language --, or hybrid -- the environment consists of a Mixed Reality, agents interact both with humans in natural language and with other agents using an agent communication languages.
The last option is the most challenging, and the most interesting for many purposes including entertainment, training, and agent-based modeling and simulation.
The literature dealing with integrating intelligent software agents, virtual reality and natural language processing is scarce and it is even scarcer when we consider a mentalistic approach to design and implement agents [1,2].
The proposed research aims at designing and developing a general purpose, scalable and flexible framework for bridging these three components, namely intelligent software agents, virtual reality and natural language processing, together, and exploiting the framework in some realistic scenario where agents interact with both a virtual and a real environment, and communicate with both humans and other agents.
In accordance with the European Commission's Ethics guidelines for trustworthy AI published on April 2019 [3], the framework must ensure transparency and explainability of the decisions made by the agents are a must, and robustness and safety must be also taken into account.
[1] Brännström, A.; Nieves, J.C. A Framework for Developing Interactive Intelligent Systems in Unity. In Proceedings of the Engineering Multi-Agent Systems (EMAS 2022); Amit Chopra, 936 J.D.; Zalila-Wenkstern, R., Eds., 2022.
[2] Andrea Gatti, Viviana Mascardi. VEsNA, a Framework for Virtual Environments via Natural Language Agents and Its Application to Factory Automation. Robotics 12(2): 46 (2023)
[3] The High-Level Expert Group on AI, Ethics guidelines for trustworthy AI, April 2019, https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai
Title: Innovative teaching methods in Computer Science
Proposer: Giorgio Delzanno, Giovanna Guerrini
Curriculum: Computer Science
Abstract:
In this PhD project, we are seeking for candidates interested in innovative teaching methods for basic and advances topics in computer science.
References:
iCoding: Immersive Coding in Unity. Manuela Chessa, Giorgio Delzanno, Davide Giovannetti, Giovanna Guerrini, Filippo Manini, Davide Miggiano, Marianna Pizzo, and Eros Viola, iLRN 2023.
Smart rogaining for computer science orientation: Manuela Chessa, Giorgio Delzanno, Angelo Ferrando, Luca Gelati, Giovanna Guerrini, Viviana Mascardi, Nicoletta Noceti, Francesca Odone and Francesca Vitali, Frontiers in Computer Science Education/Education. 2022.
Giorgio Delzanno, Luca Gelati, Giovanna Guerrini, Angela Sugliano, Daniele Traversaro: Experience-Based Training in Computer Science Education via Online Multiplayer Games on Computational Thinking. HELMeTO 2022: 459-470
Chiara Capone, Rafael H. Bordini, Viviana Mascardi, Giorgio Delzanno, Angelo Ferrando, Luca Gelati, Giovanna Guerrini:
Smart RogAgent: Where Agents and Humans Team Up. PRIMA 2019: 541-549
Giorgio Delzanno, Giovanna Guerrini, Viviana Mascardi, Daniele Traversaro: PyWeCode: Towards a Collaborative Coding Framework based on the arcade Python Library. UMAP (Adjunct Publication) 2020: 107-113
Daniele Traversaro, Giovanna Guerrini, Giorgio Delzanno: Sonic Pi for TBL Teaching Units in an Introductory Programming Course. UMAP (Adjunct Publication) 2020: 143-150
Davide Ancona, Lorenzo Benvenuto, Giorgio Delzanno, Gianluca Gambari: Flow Programming: A Flexible way to bring the Internet of Things into the Lab. UMAP (Adjunct Publication) 2020: 155-158
Chiara Capone, Rafael H. Bordini, Viviana Mascardi, Giorgio Delzanno, Angelo Ferrando, Luca Gelati, Giovanna Guerrini: Smart RogAgent: Where Agents and Humans Team Up. PRIMA 2019: 541-549
Title: Smart Wearable Data
Proposer: Giorgio Delzanno
Curriculum: Computer Science
Abstract:
Interoperable collaborative platform for wearable data acquisition and analysisWearables refer to those intelligent devices that interact directly with people through contact with their body. These technologies can be either a garment or an accessory such as a pair of glasses, a watch or a bracelet. These devices can be connected to other devices such as smartphones, through the wireless network system or via bluetooth technology thus allowing the detection, storage and exchange of vital data in an immediate and continuous way even without the need for human intervention.
As a case study, in a recent rogaining activity organized by the Liguria Region in collaboration with Edutainment Formula, MD class IIb sensors were tested by a team of 6 participants (Unige students) wearable textiles produced by Comftech S.r.L. capable of detecting the following vital parameters: ECG, heart and respiratory rate, data relating to the movement of the person through the accelerometer, well-being and stress indices based on the analysis of HRV cardiac variability. The test allowed an initial verification of the robustness of the reports produced at the end of the activity (about 2 and a half hours) through interviews with the people involved in the experiment.
The typology of sensors considered is of particular interest for various reasons. First of all, it does not require cables or other type of connections but only a t-shirt with textile sensors and the use of a smartphone connected via MD certified protocol with the device.
The person wearing the device therefore does not have to intervene during the monitoring process: an operator can freely use tools with both hands, a frail person can walk with the help of supports, a small child can interact with usual objects in the its environment, an operator with a VR headset can operate the controllers, etc.
Starting from this type of case-study, which can be extended with other motion and position data acquisition kits, the project aims to build a multi-purpose platform for the acquisition, processing and sharing of data, code and documentation with a set of users interested in both data consultation and experimentation with new algorithms and analysis methods.
To enable this possibility we would like to build an on-premise platform based on the following components:
• User registration portal according to OAuth/OAuth 2 standards.
• App to acquire data from sensors using certified protocols.
• Data acquisition node (eg InfluxDB/Kafka depending on the context) with user authentication and storage in document databases with standard Json confirmations.
• Data sharing and processing node based on Jupyter Hub to be able to make dashboards and executable documents available to users, to ensure reproducibility, through Notebook.
Jupyter Hub allows the configuration of customized Virtual Environments for different types of users with modules with the most common analysis libraries (e.g. PyTorch, TF, scikit, etc.)
• Kebana-based dashboard creation node
• Node for building multi-protocol middleware based on Node-red (open source) and professional platforms such as PTC Thingworx.
Jupyter Hub provides libraries for transforming notebooks into services with REST entry points.
They can be used to implement user-generated ingestion and processing pipelines and, thus, simplify the deployment
process of an application on-premise and on edge devices.
The resulting platform could be used as a hub for a community of researchers (academic and industrial) interested in sharing
datasets, code, results and documentation in an environment protected by authentication (Jupyter Hub is an
OAuth 2 authentication service).
The resulting system is compliant with Open Science oriented research infrastructures such as D4Science and Cyverse.
At the hardware infrastructure level, we would like to build a cluster with a 10Gb network and nodes dedicated to the
services described above (Jupyter Hub, Kebana, Node-red, Influxdb/Mongodb) with the possibility of exploiting a series of distributed computing nodes with the Apache suite Spark and a set of GPU compute devices for AI algorithms (Jetson) and an external node for Thingworx.
Possible applications of the system are envisaged in various fields:
- Continuous monitoring of (groups of) frail people (e.g. rehabilitation)
- Monitoring of teams in outdoor/indoor activities
- Monitoring of VR activities (e.g. serious games, embodiment, etc.)
- Monitoring of critical activities (e.g. drivers, etc.)
The project requires the definition of a user manual for the system and the related communication interfaces and a prototype of the acquisition platform to be used with at least 5 devices.
Title: Design and Validation of IoT/Big Data Applications
Proposer: Giorgio Delzanno
Curriculum: Computer Science
Description: Multithreaded, parallel and distributed applications are at the core of modern software architectures and applications. For instance, reference architectures for the Internet of Things are typically based on scalable distributed engines mixing the use of local, low-power system on chips (i.e. Edge and Fog computing) together with remote computing infrastructures (e.g. the Cloud) for data ingestion, real time and batch processing and data visualisation. We can find here multiple open challenges ranging from new architectures for data processing (e.g. Spark), to seamlessly provide data integration and code migration, to enable high performance/efficient computing as services (HPCaaS), exploiting the computational power of edge computing resources at their best (e.g. neural accelerators).
In this general setting we are interested in the more specific research lines in the context of the PNRR Ecosystem "RAISE" project:
- IoT/Big data applications for smart cities/ports/building combining edge and cloud computing [1,2]
- Validation of IoT applications using run time verification in combination with process mining [3] and parameterized verification [4,5,6]
Link to the group or personal webpage: Giorgio-Delzanno's-webpage DBLP
References:
[1] L. Bixio, Giorgio Delzanno, S. Rebora, M. Rulli:
A Flexible IoT Stream Processing Architecture Based on Microservices. Inf. 11(12): 565 (2020)
[2] A Solution for Improving Robustness of GNSS Positioning from Android Devices,
Lorenzo Benvenuto's Phd Thesis in Computer Science and Systems Engineering, University of Genoa, May 2022
Supervisors: Giorgio Delzanno (Dibris) and Tiziano Cosso (Gter)
[3] D. Ancona, L. Benvenuto, G. Delzanno, G. Gambari:
Flow Programming: A Flexible way to bring the Internet of Things into the Lab.
UMAP 2020: 155-158
[4] Angelo Ferrando, Giorgio Delzanno:
Incrementally Predictive Runtime Verification.
CILC 2021: 92-106
[5] S. Conchon, G. Delzanno, A. Ferrando:
Declarative Parameterized Verification of Distributed Protocols via the Cubicle Model Checker.
Fundam. Informaticae 178(4): 347-378 (2021)
[6] Sylvain Conchon, Giorgio Delzanno, Arnaud Sangnier:
Verification of Contact Tracing Protocols via SMT-based Model Checking and Counting Abstraction.
CILC 2021: 77-91
Supervisor(s):Maurizio Leotta, Filippo Ricca
Title: Developing Novel Test Automation Solutions for Web and Mobile Applications
Keywords: End-to-end Testing, Test Automation, Software Engineering
Curriculum: Computer Science
Abstract:
Testing web and smartphone apps can take a long time, both for the complexity of these products and for the variety of environments through which end users can use them. On the other hand, the need to reduce the distribution times of new versions and the progressive adoption of Agile methodologies in software development lead to ever smaller margins to guarantee the effective quality of the final product. In this context, the creation of automated tests becomes an essential requirement to increase efficiency and quality while at the same time reducing the overall costs.
The objectives/steps of the PhD are:
1) Selecting one of the possible interesting topics in the context of web or mobile testing, for example: automated test suite generation, test suite fragility reduction, test suite execution optimization, test suite flakiness reduction etc.
2) Studying the state of the art in this specific topic
3) Devising one or more solutions/algorithms advancing the state of the art, and implementing them.
4) Executing empirical studies comparing existing solutions with the novel proposals.
https://www.disi.unige.it/person/LeottaM/
https://www.disi.unige.it/person/RiccaF/
References
External collaborations:
Both from Academia (e.g., University of Lugano, Blekinge Institute of Technology, University of Madrid) and local Industries.
Title: Detection of the origin of movement: Data processing and developments software libraries
Supervisors: Antonio Camurri, Giorgio Gnecco (IMT Lucca), Marcello Sanguineti, Gualtiero Volpe
Keywords: Origin of movement, Human-Computer Interaction, Affective computing, Motion capture, Multimodal interfaces and systems.
Curriculum: Computer Science
Abstract. In the thesis, the approach proposed in [1,2] for the analysis of the origin of human movement will be developed in from the points of its exploitation in real-time contexts, with particular emphasis on data processing and production of software libraries to allow its real-time application. The perceived origin of movement is the point at which a movement appears to originate from the point of view of an observer. The importance of full-body movements in conveying affective expressions and social signals is widely recognized by the scientific community [3,4], and a growing number of applications exploiting full-body expressive movement and non-verbal social signals are available. The possibility to automatically measure movement qualities such as its origin demonstrated to be very important in many different interactive applications, including therapy and rehabilitation in autism, and in cognitive and motor disabilities [5]. In [1,2], an approach based on a mathematical model called cooperative game is developed to study the origin of movement. A mathematical game is built over a graph structure representing the human body and a utility function related with movement features was defined. The games of the players are the joints of the human body. An attribute of the players, called “Shapley value” is evaluated and used to study movement. The targets of the proposed thesis are two. The first one consists in refining the methodology developed in [1,2] by considering a larger set of movement features, namely speed, tangential acceleration, and angular momentum. It will be investigated which feature is best at predicting the origin of movement. The method will be applied to a data set of Motion Capture data of subjects performing expressive movements, also by applying suitable filtering techniques to such data. The second target is the development of software libraries for the analysis of the origin of movement. The departure point is a software, written in Matlab, which extracts movement features (speed, tangential acceleration, kinetic energy, angular momentum) for two skeleton models of the human body, which refer to two different spatial scales. Such features are filtered and combined to compute a dissimilarity measure for the joints, from which a transferable utility game on an auxiliary graph is constructed. Finally, the vector of Shapley values for that game is computed (for the case in which the Shapley value coincides with the weighted degree centrality), and normalized with respect to the maximum Shapley value. The software allows for the possibility of computing the Kendall correlation between different rankings of joints. This software should be re-written in a suitable language and optimized from the point of view of efficiency and runtime, in such a way to make it exploitable for real-time analysis of the origin of movement.
External collaborations. This thesis will benefit from the ongoing activities of the European-funded FET PROACTIVE EnTimeMent 4-year project (entimement.dibris.unige.it). EnTimeMent aims at the foundation and consolidation of radically new models and motion analysis technologies for automated prediction and analysis of human movement qualities, entrainment, and non-verbal full-body social emotions. The approach is grounded on novel neuroscientific, biomechanical, psychological, and computational evidence dynamically suited to the human time, towards time-adaptive technologies operating at multiple time scales in a multi-layered approach. The research will benefit also from the motion capture and multimodal technology infrastructure available at Casa Paganini-InfoMus of Dibris (www.casapaganini.org). Specific application testbeds to validate and evaluate research results will be identified in one of the EnTimeMent scenarios (cognitive-motor rehabilitation, performing arts, sport). The thesis will also benefit from the ongoing activities of the Università Italo-Francese project GALILEO 2021 no. G21 89, “Automatic movement analysis techniques for applications in cognitive/motor rehabilitation”, between Università di Genova, IMT - Scuola Alti Studi Lucca, and Université de Montpellier.
Research activities will include collaborations and short residencies at the premises of one or more partners of the EnTimeMent project, including Qualisys (motion capture industry), Euromov – University of Montpellier (Prof Benoit Bardy), UCL University College London (Prof Nadia Berthouze), with IMT - School for Advanced Studies, Lucca (Prof. Giorgio Gnecco), and with the incubators of startups GDI Hub (London) and Wylab (Chiavari).
Link to the group/personal webpage:
entimement.dibris.unige.it
References
[1] K. Kolykhalova, G. Gnecco, M. Sanguineti, G. Volpe, A. Camurri: Automated Analysis of the Origin of Movement: An Approach Based on Cooperative Games on Graphs. IEEE Trans. on Human-Machine Systems, Vol. 50, pp. 550-560, 2020.
[2] O. Matthiopoulou, B. Bardy, G. Gnecco, D. Mottet, M. Sanguineti, A. Camurri: A computational method to automatically detect the perceived origin of full-body human movement and its propagation. Proc. Multi-Scale Movement Technologies ACM-ICMI 2020 Int. Workshop. 25-29 Oct. 2020, pp. 449-453.
[3] B. De Gelder, “Why bodies? Twelve reasons for including bodily expressions in affective neuroscience,” Philosophical Transactions of the Royal Society B: Biological Sciences, vol. 364, no. 1535, pp. 3475–3484, 2009.
[4] A. Kleinsmith, N. Bianchi-Berthouze: Affective body expression perception and recognition: A survey,” IEEE Transactions on Affective Computing, vol. 4, no. 1, pp. 15–33, 2013.
[5] S. Piana, A. Staglianò, F. Odone, A. Camurri: Adaptive Body Gesture Representation for Automatic Emotion Recognition. New York, NY, USA: ACM, Mar 2016, vol. 6, no. 1, pp. 6:1–6:31. Available: http://doi.acm.org/10.1145/2818740
Title: Automated analysis of expressive qualities of full-body human movement
Supervisors: Antonio Camurri, Giorgio Gnecco (IMT Lucca), Marcello Sanguineti, Gualtiero Volpe
Curriculum: Computer Science
Keywords: Affective computing, Human-Computer Interaction, Motion capture, Multimodal interfaces and systems, Operation research, Games.
Abstract. The role played by full-body movements in conveying affective expressions and social signals is widely recognized by the scientific community [1], and a growing number of applications exploiting full-body expressive movement and non-verbal social signals is available. The possibility to automatically measure movement qualities is very valuable in many different interactive applications, including therapy and rehabilitation in autism, and in cognitive and motor disabilities. In [2,3], the automated analysis of the origin of movement (i.e., where in the body the movement initiates), which is an important component in understanding and modelling expressive movement, was investigated.
In the thesis, the approach proposed in [2,3,4,5] will be developed in several directions by using the same approach, based on a mathematical model called cooperative game. In general, mathematical games [6] study interactions among subjects, by modelling conflict or cooperation between intelligent entities called players. In the analysis of full-body human movement, the game model is built over a suitably-defined three-dimensional structure representing the human body. The players represent a subset of body joints. Each group of players has an associated utility, which represents their joint contribution to a common task. Using a utility constructed starting from a movement-related feature such as speed, a cooperative game index called Shapley value [6] can be exploited to analyse expressive qualities (e.g., to identify the movement origin, as done in [2,3,4,5] using the feature speed). Targets of the proposed thesis include, e.g:
- Considering different formulations of the cooperative game and/or different cooperative indices.
- Extracting and embedding into the model other features and/or sets of features calculated from movements, such as position, acceleration, speed, jerks, and angular acceleration.
- Investigating the time series of the Shapley values to capture the dynamics of movement in finer details (e.g., the importance of different timescales in recognizing a specific movement).
- Modelling biomechanical constraints, which determine the way we move as well as the way we perceive movements.
- Analysing the automatic detection of movement qualities different from the origin of movement.
- Conceiving novel experiments, in order to build up the movement repertoire and enlarge the available motion-capture data set.
As an outcome of the thesis, a larger set of computational methods and software tools will be available for the automatic analysis of expressive qualities associated with full-body human movement.
External collaborations. This thesis will benefit from the ongoing activities of the European-funded FET PROACTIVE EnTimeMent 4-year project (entimement.dibris.unige.it). EnTimeMent aims at the foundation and consolidation of radically new models and motion analysis technologies for automated prediction and analysis of human movement qualities, entrainment, and non-verbal full-body social emotions. The approach is grounded on novel neuroscientific, biomechanical, psychological, and computational evidence dynamically suited to the human time, towards time-adaptive technologies operating at multiple time scales in a multi-layered approach. The research will benefit also from the motion capture and multimodal technology infrastructure available at Casa Paganini-InfoMus of Dibris (www.casapaganini.org). Specific application testbeds to validate and evaluate research results will be identified in one of the EnTimeMent scenarios (cognitive-motor rehabilitation, performing arts, sport). The thesis will also benefit from the ongoing activities of the of the Università Italo-Francese project GALILEO 2021 no. G21 89, “Automatic movement analysis techniques for applications in cognitive/motor rehabilitation”, between Università di Genova, IMT - Scuola Alti Studi Lucca, and Université de Montpellier.
Research activities will include collaborations and short residencies at the premises of one or more partners of the EnTimeMent project, including Qualisys (motion capture industry), Euromov – University of Montpellier (Prof Benoit Bardy), UCL University College London (Prof Nadia Berthouze), with IMT - School for Advanced Studies, Lucca (Prof. Giorgio Gnecco), and with the incubators of startups GDI Hub (London) and Wylab (Chiavari).
Link to the group/personal webpage:
entimement.dibris.unige.it
References
[1] A. Kleinsmith and N. Bianchi-Berthouze, “Affective body expression perception and recognition: A survey,” IEEE Transactions on Affective Computing, vol. 4, no. 1, pp. 15–33, 2013.
[2] K. Kolykhalova, G. Gnecco, M. Sanguineti, A. Camurri, and G. Volpe: Graph-restricted game approach for investigating human movement qualities”. Proc. 4th Int. Conf. on Movement Computing (MOCO ’17). London, UK: ACM, 2017, article no. 30, 4 pages.
[3] K. Kolykhalova, G. Gnecco, M. Sanguineti, G. Volpe, A. Camurri: Automated Analysis of the Origin of Movement: An Approach Based on Cooperative Games on Graphs. IEEE Trans. on Human-Machine Systems,. Vol. 50, pp. 550-560, 2020.
[4] O. Matthiopoulou, B. Bardy, G. Gnecco, D. Mottet, M. Sanguineti, and A. Camurri: A computational method to automatically detect the perceived origin of full-body human movement and its propagation. Proc. Multi-Scale Movement Technologies ACM-ICMI 2020 Int. Workshop. 25-29 Oct. 2020, pp. 449-453.
[5] O. Matthiopoulou, B. Bardy, A. Camurri, G. Gnecco, D. Mottet, and M. Sanguineti: Detection of the Origin of Movement: Graph-Theoretical Model and Data Processing. Int. Conf. on Optimization a nd Decision Science. Firenze, 30 Aug.-2 Setp 2022.
[6] M. Maschler, E. Solan, and S. Zamir, Game Theory. Cambridge, UK: Cambridge University Press, 2013.