2026 Call

Requirements: students enrolled at either the first with at least 20 CFUs and a GPA of at least 27.5 or second year of the MSc track, with 50 CFUs and a GPA of at least 28

Research topics in Automation and Control Engineering

Proposer: Matteo Corno, Lorenzo Fagiano, Simone Formentin, Luigi Piroddi, Maria Prandini, Sergio Savaresi, Mara Tanelli, Gian Paolo Incremona, Alessio La Bella Topic: Theory and application of control systems The research area includes all the topics related to the theory and application of modeling, control, identification and learning methods for dynamic systems. Applications include, but are not limited to, energy systems, smart grids, vehicles, finance. Compatible tracks: ATM, CSE Positions available: 3

Proposer: Renato Casagrandi, Lorenzo Mari, Maria Pradini, Simone Garatti, Fabio Dercole, Carlo Piccardi Topic: Modelling, optimization, and control of complex systems The research area includes all topics related to modelling, optimization and control for complex systems possibly interacting over a network or involving hybrid dynamics. Examples include social, biological, epidemiological, transportation and power networks. Compatible tracks: ATM, CSE Positions available: 3

Research topics in Electric Engineering

Proposer: Luca Di Rienzo Topic: Electromagnetic modeling of submarine power cables The transition towards renewable energy sources is tightly tied to the capability to transmit and dispatch energy from the production places to where it is consumed. Within this framework, underground and submarine cables for power transmission and distribution are an essential component of the global energy transition. The proper design of these power cables must take into account several different and often contrasting requirements and constraints and their accurate electromagnetic modeling is crucial. Computation of electric and magnetic losses, modeling of the electric field in the insulation material, and evaluation of the per unit length parameters are challenging tasks. The research work will contribute to the numerical implementation of new formulations that can model the involved electromagnetic phenomena.

Proposer: Flavia Grassi Topic: Prediction models and test procedures to assure electromagnetic compatibility in the automotive sector Assuring electromagnetic compatibility (EMC) performance of electronic components and communications systems installed on board hybrid and electric vehicles represents a great challenge for electrical/electronics engineers. For this reason, the development of prediction models and test procedures to provide reliable estimates (also in statistical terms) of the expected levels of interference is desirable in order to support the design of electrical and electronic components and subsystems starting from the early design stages. The research activity that will be proposed to the student will contribute to the research activities currently carried out by the EMC group in this field, and will be focused more on modeling/simulation rather than on experimental activity depending on the specific student interests.

Research topics in Electronic Engineering

Proposers: Giuseppe Bertuccio Topic: Semiconductor Radiation Detectors and Integrated Circuits for Scientific Applications Semiconductor Radiation Detectors (SRD) are widely employed in fundamental sciences and have numerous scientific, medical and industrial applications. The continuous increase in complexity and performance of new SRD requires advances in Application Specific Integrated Circuits (ASIC) for signal readout and processing. The research activity can be directed through one of these two paths: 1) the theoretical and experimental study of the most advanced semiconductor radiation detectors in Silicon (such as Silicon Drift Detectors and Low-Gain Avalanche Diodes) and in Compound Semiconductors (GaAs, CdTe, CdZnTe, SiC) or 2) the design of low-noise and low-power CMOS ASIC for SRD. Both topics are oriented to applications toward Synchrotron X-ray sources and Space Telescopes for Astrophysics or non-destructive material analysis in Industries. The research activities are made in collaboration with the National Institute for Nuclear Physics (INFN), the National Institute for Astrophysics (INAF), the Italian Space Agency (ASI), the European Space Agency (ESA), Elettra Synchrotron in Trieste and several industrial partners.

Proposers: Marco Carminati, Carlo Fiorini Topic: Development of innovative detection systems and readout ASICs for fundamental physics and medical imaging applications The student will be involved in stimulating team activity carried out within international collaborations and concerning the development of innovative detection systems and readout ASICs to be used in experiments for both fundamental physics (nuclear physics, neutrinos search, X-ray astronomy) and applied physics (analytical instrumentation for material analysis, industrial applications). Projects of the team regard also the development of sensors and electronics for medical imaging applications. During the program, the student will have the opportunity to contribute to the development of a broad range of components, circuits and systems, from radiation sensors, to low-noise integrated circuits, to data acquisition systems, to a complete instrument ans its use in the target application. Positions available: 2

Proposers: Andrea Castoldi Topic: Developement of novel Germanium Drift Detectors The study and development of Germanium detectors with 'drift' topology is a research topic that can give a break-through in the field of imaging and spectroscopy of X- and gamma-rays. Germanium is one of the purest semiconductor crystals available today with excellent carrier transport properties and high detection efficiency of X and gamma rays that makes it the ideal candidate for such task. The design of novel 'drift' topologies, i.e. based on 3D shaping of the electric field inside the semiconductor volume, combined with the unique material properties has the potential of unprecedented low-noise performance on a wide energy range. The student will be involved in the team work in the development of prototypes of Germanium drift detectors and of the low-noise readout electronics with a broad range of activities from device simulation, experimental characterization as well as data analysis.

Proposers: Giorgio Ferrari, Marco Sampietro Topic: Neuromorphic hardware for an energy-efficient brain-inspired computation Many recent successes of artificial intelligence have been achieved using artificial neural networks that are loosely inspired by the brain's functioning and implemented as software on high-performance computers with a traditional architecture. However, their energy-intensive nature hinders their use in portable and compact autonomous systems. To overcome this challenge, the research explores radically different hardware architectures inspired by biological models, particularly focusing on energy-efficient neuromorphic systems employing physical neurons for computation. In this context, the research at the I3N laboratory of the Politecnico di Milano is dedicated to developing neuromorphic circuits in CMOS technology. These circuits feature neurons and synapses implemented through compact analog circuits, targeting very low power consumption and long-term memory. The research encompasses both the electronic design of the chip and the learning methodologies more suitable for these neural networks, taking into account the constraints imposed by their physical implementation and exploring potential enhancements given by the electronic noise inherent in analog approaches.

Proposers: Chiara Guazzoni Topic: Development of innovative high-rate rad-hard detection systems for radioactive and stable beams Development of innovative high-rate detection systems for radioactive and stable beams The activity proposed to the candidates frames in the upgrade of the superconducting cyclotron of INFN (Istituto Nazionale di Fisica Nucleare), at Laboratori Nazionali del Sud, in Catania (Italy) towards high-luminosity radioactive beams for nuclear physics experiments. Politecnico di Milano research staff, in collaboration with national and international partners, is in charge of the development of a high-rate radiation-tolerant tagging system, equipped with a dedicated frontend. The research project is on-going on and will last till 2028, therefore there is adequate time span for a M.Sc. and also a following PhD activity or just a M.Sc activity. The student(s) will collaborate in the detector qualification, frontend design and qualification, interconnection and mechanical assembly design and in the full detection system modelling. Last but not least the student(s) will take part in the beam tests in different national and international facilities. Aside the experimental activity a full position is available dedicated to detector simulation. The system will also have biomedical interest as it will be used to qualify radioactive beams for combined diagnosis and treatment in radiotherapy with tests carried out also at the Italian National Centre for Adrotherapy (CNAO in Pavia). The students (up to 3) will have the possibility to shape the focus of the research activity according to their skills and interests and to work independently on different aspects of the project or jointly collaborate. The student(s) will experience the working atmosphere of a real research lab, with hands-on approach combined with a strong theoretical background in the field of radiation detectors and low-noise frontend electronics, sharing the daily lab life with PhD students, junior and experienced researchers.

Proposers: Chiara Guazzoni Topic: Upgrade of the FARCOS detection system for nuclear safeguards and security The activity proposed to the candidates frames in the upgrade of a novel high-resolution detection system, named FARCOS and originally built for nuclear physics applications, in view of its upgrade as smart module for nuclear safeguards and security. The project is suitable for 1 or 2 people working together. The idea is to build a standalone module, equipped with all the required electronics (frontend and backend) in a compact module, and to develop the proper control software and user interface. In addition, the project foresees the optimization of the algorithms used for the identification of the detected particles and for the gamma ray discrimination. The student(s) will have the possibility to shape the focus of their research activity according to their skills and interests and to work independently on different aspects of the project or jointly collaborate. In addition, the student(s) will experience the working atmosphere of a real research lab, with hands-on approach combined with a strong theoretical background in the field of radiation detectors and low-noise frontend electronics, sharing the daily lab life with PhD students, junior and experienced researchers. Last but not least they will have the chance to develop a full detection module and to operate it!

Proposers: Angelo Gulinatti, Ivan Rech Topic: Single-photon detectors and applications Recently, the field of single-photon detection has experienced an exponential growth due to the emergence of a large number of scientific and industrial applications that require the capability of detecting the light down to the single-photon level. Among them are laser ranging, single-molecules analysis, and optical quantum computing, which are regarded as key enabling technologies to address some of today?s most challenging problems like the development of new drugs, the study of climatic changes and the autonomous driving. The Single-Photon Avalanche Diode (SPAD) has emerged as the detector of choice for these applications thanks to its remarkable performance. However, many progresses are still needed to bring these applications in everyday life. The research activity is aimed at addressing these limitations by developing innovative solutions for the detection of single photons with unprecedented capabilities in terms of photon detections efficiency, timing jitter, spectral sensitivity, and integrability into large arrays of detectors. The activity will involve one or more of these fields: design and simulation of new detector structures, study and modeling of unexplored detector properties, development of new experimental characterization techniques, design of new front-end circuitry, development of application specific architectures. Positions available: 2

Proposers: Daniele Ielmini, Christian Monzio Compagnoni, Alessandro Spinelli Topic: Emerging memory devices and in-memory computing The future evolution of integrated electronictechnologies will see a closer interaction of logic and memory devices, overcoming the traditional design schemes in the attempt to constantly reduce power consumptions and improve system performance. This is today triggering the research both on new, fast and energy-efficient memory devices that could be integrated directly in logic processes and on new computational forms that could be directly performed inside memory arrays. The proposed projects for the Honours Program will make contributions to these two research lines. In the former, activities will be focused on the experimental, numerical and theoretical investigation of the basic physics, the working principles and the performance of emerging memory devices, such as nanoscale ferroelectric memories and magnetoresistive memories. In the second, in-memory computing will be investigated both theoretically and experimentally, focusing on hardware neuromorphic systems exploiting different types of memory arrays, e.g., floating-gate, resistive and phase-change arrays, as artificial synaptic arrays. The results of the projects will pave the way to the future evolution of micro and nanoelectronics. Positions available: 3

Proposers: Dario Natali Topic: Plastic Electronics Organic semiconductors are carbon-based compounds that can be conveniently processed from solution at almost room temperature. As a consequence, it is possible to manufacture devices, circuits and systems by means of non-lithographic techniques, such as inkjet printing, screen printing, flexography to cite but a few. The benefits of printed electronics include: low cost, mechanical flexibility, ease of production and integration, and the possibility of addressing a variety of non-conventional applications: flexible displays, image sensors, bio-sensors, smart labels, RF-ID tags, edible electronics, wearable electronics, and more. The research encompasses all the aspects, issues and challenges of plastic electronics: the fundamentals of charge carrier transport related to the energetically disordered landscape sampled by carriers; device modeling and simulation through innovative ad-hoc developed codes; printing process engineering; logic, analog, optoelectronic circuits and systems design, realization and characterization.

Proposers: Marco Sampietro, Francesco Zanetto Topic: Integrated electronic systems for the control of optical computers Manipulation of light in waveguides produced on a silicon layer with the same microelectronic technology used for electronic chips opens the realm of optical computing. For the scenario to be realistic, many difficulties should be overcome, the principal being the strong thermoptic coefficient of Silicon that prevent Silicon Photonics to be operated in open loop. Our research targets this aspect by designing and operating electronic systems for the control of Photonic chips in a close loop configuration. To this goal non-invasive monitoring devices to inspect the light inside optical waveguides will be developed as well as integrated electronic architectures that detect these signals and perform feedback control. The electronics will be mixed-signal integrated chips tailored to the photonic chips for the stabilization and setting of its operation. Flip-chip mounting will be explored as well as direct integration of electronic functionalities into the same photonic chip, in a truly integrated optical computer system.

Proposers: Alberto Tosi, Federica Villa, Franco Zappa Topic: Microelectronics, devices, and electronic systems for LiDAR, quantum physics and bioimaging 1. "ELECTRON DEVICES: design, simulation, and fabrication of single-photon detectors in Si, InGaAs/InP and Ge". The research aims at conceiving new microelectronic sensors for detecting single photons in the 200nm-1,700nm range, with picosecond-timing precision. Different semiconductors (Si, Ge, InGaAs, InP, GaAs) and novel processing (planar, three-dimensional stacking, wafer-to-wafer bonding, vertically-grown micro-crystals) will be pioneered. Funding from European Commission “Horizon 2020”, “Horizon Europe” and “European Defense Fund” programmes, European Space Agency (ESA), USA DARPA agency, national funds. 2. MICROELECTRONICS: schematics, layout, fabrication and validation of ASIC chips for systems-on-chips". The research aims at integrating detector arrays, in-pixel mixed-signal electronics and on-chip digital processing into one single chip, for imagers providing 2D movies at high (>100,000 frames/s) rates, single-photon sensitivity, time resolved imaging for LiDAR and functional bioimaging, and photon coincidence for quantum physics applications. Different microelectronic technologies (CMOS, BCD, Si-Ge and 3D-stacked) and foundries will be exploited. Funding from “Horizon2020”, “Horizon Europe” and “European Defense Fund” EC programmes, NATO “Safety for Peace” programme, European Space Agency (ESA), JPL-NASA agency. 3. "ELECTRONIC SYSTEMS: development and system integration of advanced electronic systems for sensing and imaging". The research aims at prototyping electronic products with FPGA, µC, and DSP devices, for different applications (e.g., time-of-flight 3D automotive vision, quantum optics experiments, quantum computing chips, non-invasive biomedical parameter acquisitions). Various methodologies, hw/fw/sw co-partitions, devices (Xilinx, Altera, ARM-Cortex) and high-level languages will be exploited. Funding from JPL-NASA agency, European Southern Observatory, and European Space Agency). Positions available: 3

Research topics in Computer Science and Engineering

Proposers: Giovanni Agosta Topic: Compiler technology for Quantum Logic Synthesis Quantum computers promise to revolutionise the landscape of computing, e.g. enabling the breaking of classical public key cryptography. However, designing quantum algorithms and integrating them with classical ones is a complex task. MLIR provides a versatile compiler framework that can be adapted to this task. The goal of the thesis is, starting from an MLIR-based intermediate representation of quantum circuits, develop transformations to optimize the mapping of logical circuits to quantum computers.

Proposers: Francesco Amigoni Topic: Coordinated Path Planning for Multi-Robot Systems Teams of cooperative agents can provide effective and efficient solutions for tasks in warehouse environments as well as in applications such as environmental monitoring, search and rescue, and patrolling. However, such multi-robot systems pose significant challenges for the development of planning methods, based on classical or learning techniques, that are required to accomplish tasks with an appropriate level of autonomy.

Proposers: Salvatore Andolina Topic: Human-Centered AI-Augmented Search Large language models offer powerful ways to find and use information, but raise concerns about accuracy, bias, and overreliance. This thesis designs AI-augmented search systems that improve effectiveness while promoting critical thinking, transparency, controllability, and trust.

Proposer: Danilo Ardagna Topic: AI Application Management through Reinforcement Learning in Computing Continua Reinforcement learning (RL) is promising to optimize AI applications running in edge systems. However, RL agents perform poorly during initial exploratory interactions with the environment.T he project goal is to develop a new framework leveraging prior knowledge to address this issue.

Proposers: Danilo Ardagna Topic: Deep Dive in Agentic AI GenAI is expected to significantly impact the global economy, with projections showing between US$2.6-4.4 trillion in annual GDP by 2030. The future of GenAI is agentic, where AI agents collaborate in real-time to automate complex tasks. This thesis compares major Agentic AI frameworks based on reasoning, adaptability, and computing demands evaluating their efficiency and suitability for various applications.

Proposer: Federica Arrigoni Topic: 3D Reconstruction from Images Under a Simplified Camera Model A relevant problem in Computer Vision is 3D reconstruction from images, also known as "Structure from Motion". Typically, a static scene is considered (like a historical monument), which is observed from multiple cameras at different locations and viewpoints, generating the input images. The goal of this project is to develop novel optimization-based methods to perform 3D reconstruction under alternative camera models. Particular focus will be given to the "affine camera", which represents a simplified model that works well under certain circumstances (for example, when the depth of the scene is nearly constant relative to the cameras). This approximation results in linear systems (instead of polynomials employed by the classical pinhole camera), which are much easier to manage.

Proposers: Luciano Baresi Topic: PyCITO Class Integration Test Order (CITO) generation is a key problem in object-oriented software testing because classes in a codebase are not independent. The order in which automated testing systems integrate and test classes can strongly affect the duration, cost, and effectiveness of the overall testing campaign. Early approaches to CITO mainly relied on simple rule-based techniques, whereas more recent work has begun to explore Reinforcement Learning (RL). Although these initial results are promising, they focus primarily on Java systems and do not fully leverage the potential of modern RL methods. This work aims to advance the state of the art by exploring more powerful RL algorithms and extending the problem to additional programming languages, with particular attention to Python and its specific characteristics. A further goal is to investigate how these techniques can be integrated into modern CI/CD workflows.

Proposer: Anna Bernasconi Topic: Warning system for viral genomic surveillance Unprecedented availability of viral genomes on public databases makes viruses' evolution exploration possible with data-driven methods. The proposed research will develop data science workflows to support surveillance of viral species (e.g., SARS-CoV-2, Monkeypox, Avian Influenza).

Proposer: Andrea Bonarini Topic: Physical metaverse We would like to explore the limits of neuroplasticity to map sensor data, collected in the real world by a robotic avatar, into a virtual world based on VR and haptic devices. The aim is to reach the embodiment of the subject into a body with different interaction possibilities. The work is part of a multidisciplinary project including researchers and students from Design, Art and Psychology. Additional team members: Federico Espositi

Proposers: Cristiana Bolchini Topic: Design and Analysis of Resilient AI/ML Applications Against Hardware Faults This research focuses on methods and tools to evaluate the resilience of AI/ML-based applications against faults in the underlying hardware platform running the application in the context of safety-related contexts (e.g., automotive, health, space). Additional team members: Luca Cassano, Antonio Miele

Proposers: Giacomo Boracchi Topic: Deep Learning for Emerging Imaging Sensors A generation of revolutionary sensors (Single Photon Avalanche Detectors, SPAD) can measure each photon hitting a pixel, recording streams. CNNs and ViT can't leverage SPAD's huge potential in high-framerate, low-light, or high-dynamic vision. The research addresses the design of new deep-learning models for visual recognition on SPAD. Additional team members: Edoardo Peretti

Proposers: Mark Carman Topic: Investigations in Efficiency when Fine-Tuning Large Language Models (LLMs) Large Language Models have shown amazing performance across all tasks in NLP. Recently techniques have been demonstrated for improving the reasoning capabilities of these models though reinforcement learning techniques. In this project we will investigate two directions for further improving training efficiency of LLMs: Modifying the loss function: is Reinforcement learning really needed or is self-training sufficient? Modifying the architecture: can we sparsify self-attention layers and achieve faster inference and training speeds?

Proposers: Luca Cassano Topic: Designing Secure RISC-V Microprocessors Modern processors are exposed to a number of vulnerabilities such as hardware Trojan horses and microarchitectural side-channel attacks (Spectre and Meltdown). The thesis will develop intelligent checkers to enable security into systems-on-chip.

Proposer: Matteo Castiglioni Topic: Online Learning in Economic Settings Online learning is a powerful tool for the design of algorithms for sequential decision making. Recently, it has been successfully applied to many economic settings on the Web, such as advertising, information markets, and online auctions. This thesis aims at exploring new frontiers of online learning applications in such settings.

Proposers: Davide Conficconi Topic: Fast Prototyping of Domain-Specific Heterogeneous Systems Computing systems are becoming heterogeneous and domain-specialized as ever before. This project aim at devising a fast-prototyping framework of domain-specific heterogeneous systems. This will unleash unprecedent HW/SW co-design explorations.

Proposers: Davide Conficconi Topic: NPU-GPU Heterogeneous Systems Acceleration Modern heterogeneous systems merge the latest NPUs and GPUs, offering high performance at affordable costs. This project accelerates compute-intensive tasks, partitioning them across the two architectures and optimizing communication between them.

Proposers: Paolo Cremonesi Topic: Machine Learning for Quantum Computing The speedups offered by Quantum Computing (QC) promise to reshape high-performance computing. However, QC is a nascent technology with many open challenges. This research aims to investigate how Machine Learning methods can be integrated with QC to help mitigate each other's limitations.

Proposer: Pietro Crovari Topic: Data Science ChatBot The project combines LLM, human-in-the-loop and interactive ML to create a human-computer interface, for trustworthy data analyses, involving users in requirements elicitation and decision process, benefiting users with limited computational skills. Additional team members: Pietro Pinoli, Franca Garzotto

Proposers: Maurizio Ferrari Dacrema Topic: Quantum Circuit Modeling In gate-based Quantum Computing, creating an effective quantum circuit for a given task often relies on heuristics. This project aims to develop a machine learning model to generate new quantum circuits and integrate it in methods relying on reinforcement learning, generative flow networks etc…

Proposers: Nicoletta Di Blas Topic: Response-Driven Learning with AI Current education systems treat student responses as outcomes rather than drivers of instruction. This project explores how AI can make students’ interpretative activity visible and operational in real time, enabling response-driven learning environments. The goal is to design and study systems where teaching dynamically adapts to student responses, investigating the shift from sequential to response-based instruction.

Proposer: Leandro Fiorin Topic: In-Memory Computing for Deep Learning Accelerators Digital In-Memory Computing (D-IMC) is a new paradigm to improve performance and energy efficiency of DL workloads by reducing the processor-to-memory data traffic. The thesis investigates the problem of system-level modeling and design space exploration of hardware accelerators based on D-IMC.

Proposer: Franca Garzotto Topic: Large Language Models (LLMs) and Generative AI (GenAI) for persons with cognitive impairments LLMs and and GenAI offer immense potential but pose a digital divide risk, one that separates those proficient in safely harnessing these technologies from those who cannot due to factors such as cognitive impairments. This thesis explores how LLMs and GenAI can aid the cognitively impaired. It develops innovative techniques and platforms for generating tailored, accessible conversations and multimedia content and for creating inclusive multimodal applications that enhance autonomy and learning. Additional team members: Mattia Gianotti, Pietro Crovari

Proposer: Gianmarco Genalti Topic: Active Ranking from 1-bit Feedback In real-world scenarios, we are often required to rank a set of candidates, and the only available observation is a binary outcome: pass or fail (1-bit feedback). The goal is to formalize this Active Ranking problem and derive theoretical guarantees. Additional team members: Alberto Maria Metelli, Marco Mussi

Proposers: Daniele Loiacono Topic: Artificial Creativity By combining generative AI models (such as LLMs) with interactive evolutionary computation, this research aims to create mixed-initiative tools that guide and expand human creativity across domains like games, art, music, and product design, enabling novel co-creation experiences.

Proposer: Luca Magri Topic: 3D Vision Beyond Standard Visual Cues The project investigates 3D computer vision techniques that go beyond traditional point correspondences, leveraging alternative cues such as silhouettes, object-level information, and multimodal data for robust geometric inference and 3D understanding.

Proposer: Alberto Marchesi Topic: Multi-Agent Learning Recently, machine learning techniques for multi-agent settings have received a growing attention. This lead to several breakthroughs in AI research, such as the development of superhuman bots for games like poker, go, and Starcraft. This thesis aims at designing multi-agent learning algorithms in new settings, with a particular focus on convergence guarantees and computationally efficiency.

Proposer: Alessandro Margara Topic: Efficient Execution of Data-Intensive Applications in the Compute Continuum Today, most software applications are data-intensive: they need to manage, analyze, and distribute large volumes of data in large-scale environments. The goal of this thesis is to contribute to the development of an efficient and scalable platform for the execution of data-intensive applications in Cloud-to-Edge Compute Continuum environments. Additional team members: Gianpaolo Cugola, Luca De Martini

Proposers: Davide Martinenghi, Emilia Lenzi Topic: AI for Ranking, Ranking for AI Ranking and re-ranking candidates is a fundamental problem in modern AI. This project focuses on Retrieval-Augmented Generation and LLM-based query answering, exploring advanced ranking techniques and evaluating their impact on result quality, efficiency, fairness, and distribution.

Proposer: Marco Masseroli Topic: AI and Data Science for Bioinformatics and Computational Genomics Bioinformatics research focuses on life science big heterogeneous data and information, increasingly available in many different sources, and their best integration, mining and analysis with machine learning and deep learning techniques for knowledge discovery, to better understand complex biological phenomena and improve patient stratification for best diagnosis, prognosis and clinical treatment. Additional team members: Silvia Cascianelli

Proposers: Alberto Maria Metelli Topic: Reinforcement Learning with Provable Stability Guarantees Real-world systems require stability guarantees, which are often overlooked in Reinforcement Learning (RL). This project aims to design a framework to enforce stability in RL by leveraging mixing properties of Markov chains and Lyapunov methods, with the goal of developing, analyzing, and validating new RL algorithms. Additional team members: Marco Mussi

Proposer: Antonio Miele Topic: AI/ML applications at the edge: adaptive workload distribution for performance/power optimization The research focuses on runtime AI/ML application orchestration/workload distribution to optimize resource use in the context of an ecosystem of heterogeneous edge devices and applications for smart environments.

Proposer: Luca Mottola Topic: The Battery-less Internet of Things Energy harvesting is redefining the energy constraints of traditional battery-powered IoT devices. However, such sources of energy are generally erratic, causing systems to shutdown unpredictably. We devise new software techniques to render IoT software immune to periods of energy unavailability.

Proposers: Marco Mussi Topic: Cost-Aware Reinforcement Learning from Human Feedback (RLHF) Generating and collecting human feedback to train algorithms is resource-intensive, but different types vary in their utility and difficulty. Quantifying the "cost" of a signal is crucial for designing capable agents while minimizing human burden. Additional team members: Alberto Maria Metelli

Proposer: Gianluca Palermo Topic: High Performance Virtual Screening for Drug Discovery in Urgent Computing Scenario The project wants to find solutions to accelerate the virtual screening process in drug discovery. The project considers the urgent computing scenario, similar to COVID-19, where the computation has to rely on HPC resources. Possible research directions are on code optimization strategies, exploitation of heterogeneous resources (e.g., GPUs) or quantum computers, and machine learning techniques.

Proposers: Francesco Pierri Topic: Generative AI for Science: Addressing Reproducibility and Verifiability in Research This project studies how generative AI can support scientific research while addressing challenges of reproducibility, transparency, and verifiability in existing work.

Proposer: Christian Pilato Topic: Open EDA: Scalable and Reproducible Design Automation Flows The project explores open-source EDA frameworks to enable scalable, reproducible, and customizable hardware design flows. Research includes tool development and integration, ML-driven optimization, and benchmarking on heterogeneous architectures oriented to real tape-outs.

Proposers: Pietro Pinoli Topic: Graph-Enhanced Generative AI for Healthcare Data from Text Descriptions The project will explore Generative AI models to generate realistic healthcare data from textual descriptions. The candidate will investigate how graph neural networks, conditioned on prior knowledge, enhance the quality of the generated data.

Proposers: Fabrizio Pittorino Topic: Optimization Algorithms for Continual Deep Learning In this thesis project, we will investigate optimization algorithms for deep learning with a focus on continual learning, aiming to develop robust, adaptive training strategies that mitigate catastrophic forgetting in neural networks and enhance model performance across evolving tasks. This project explores Efficient, no-waste ML scenarios, optimizing energy usage and memory footprint, with possible applications to resource-constrained environments, employing both algorithmic and hardware considerations.

Proposers: Marcello Restelli Topic: Meta Reinforcement Learning with Policy Compression for Efficient Adaptation This research explores policy compression in meta reinforcement learning to enhance adaptability and efficiency. By leveraging distillation or pruning, it aims to improve sample efficiency and transferability while reducing model complexity in diverse environments.

Proposers: Manuel Roveri, Fabrizio Pittorino Topic: Physics-Informed Neural Networks for Edge AI This thesis focuses on the design and compression of Physics-Informed Neural Networks (PINNs) for edge devices, investigating whether physical constraints enable ultra-compact and efficient models balancing accuracy and embedded resources.

Proposers: Micol Spitale Topic: Multi-Human Multi-Robot Interaction for Social Signal Computing This project explores collaborative scenarios where multiple humans and robots interact, focusing on coordination, communication, and adaptive behaviors to advance socially aware systems in complex, real-world environments.

Proposers: Monica Vitali Topic: Energy-aware design and deployment of cloud-native applications in fog computing Cloud-native applications are a collection of independent microservices that interact in a workflow. This project aims to propose a novel approach for reducing the environmental impact of cloud-native applications in distributed fog environments by designing an adaptive scheduler capable of dynamically redesigning the execution workflow considering the currently available energy mix.

Proposers: Monica Vitali Topic: Can we avoid AI to burn the planet in the LLMs era? AI's environmental impact is increasing exponentially due to the advent of LLMs and their continuous evolution in more complex systems. This project explores techniques such as prompt engineering and LLM routing towards sustainable LLMs inference.

Proposers: Monica Vitali Topic: Toward FAIR Data Preparation Pipelines for Data Sharing Data Sharing requires multiple data preparation steps to ensure quality and privacy preservation, different for different data consumers. This project explores reuse principles to reduce the computational and environmental cost of data sharing.

Proposer: Vittorio Zaccaria Topic: Security and Safety in Embedded Systems Investigate the design and implementation of secure operating systems, hypervisors, and protocols used in safety-critical environments. Subtopics could include performance and security analysis of mixed-criticality embedded systems, the use of system programming languages such as Rust and Zig in the embedded field, or formal verification of protocols.

Research topics in Telecomunication Engineering

Proposers: Mëmëdhe Ibrahimi Topic: Robust Optical-Network Optimization with Selective Hollow-Core Fiber Upgrades Hollow Core Fibers (HCF) are an advanced optical fiber technology that guides light in a hollow , i.e., air, core This thesis project will consist in formulating robust optical-network optimization problems with HCF (e.g., under uncertain traffic matrix and/or physical layer parameters), and in leveraging probabilistic Machine Learning to model the problem’s uncertainties Compatible Tracks: TLC, CSE, ATM Positions available: 1

Proposers: Massimo Tornatore, Alberto Gatto Topic: Quantum Networks Quantum key distribution (QKD) provides a method to assure telecommunication data confidentiality. However, the quantum systems provide secret key rates limited by link characteristics and time duration of the transmission. The finite-key condition could impose a strong variability to QKD connections, especially in realistic multi-node QKD networks, so a novel Routing, Channel, Key-rate and Time-slot Assignment algorithm should be developed and tested. Compatible Tracks: TLC, CSE, ELN Positions available: 1

Proposers: Paolo Martelli, Alberto Gatto Topic: Quantum Communications The proposed research activities are related to the integration of quantum key distribution in already deployed fiber optic networks and the exploitation of entanglement for secure communications. Compatible Tracks: TLC, CSE, ELN Positions available: 2

Proposers: Francesco Musumeci Topic: Converged Space-Ground Infrastructure for next-generation networking and Internet The research aims at investigating innovative networking solutions for the design and management of a Converged Space-Ground (CSG) infrastructure to support future communication services, characterized by extreme requirements in terms of latency, bandwidth, availability and security, and by the need to provision them in literally any place on Earth. Specifically, the research will focus on the integration between optical-terrestrial networks and Low Earth Orbit (LEO) satellite mega-constellations and on the modeling of resource allocation strategies in the CSG network. Compatible Tracks: TLC, CSE, ATM Positions available: 1

Proposers: Paolo Bestagini Topic: Multimedia forensics With the rise of deepfake videos, speech cloning, and AI-generated images, distinguishing real from fake content is crucial. This research explores the development of robust and explainable forensic detectors for audio, image, and video using signal processing and deep learning techniques. Compatible Tracks: TLC, CSE Positions available: 1

Proposers: Carlo Riva Topic: Coexistence of terrestrial and space communication systems Investigation and modelling of the electromagnetic interference between terrestrial (e.g. mobile networks) and space (e.g. LEO-based mega constellations) communication systems Compatible Tracks: TLC, ELN Positions available: 1

Proposers: Lorenzo Luini Topic: Advanced Earth-space communication systems Investigation of electromagnetic wave propagation oriented to the design of advanced Earth-space communication systems, which are shifting from broadcast to interactive systems, to higher frequency bands, and from GEO to LEO satellites. Compatible Tracks: TLC, ELN Positions available: 1

Proposers: Marouan Mizmizi Topic: Extreme MIMO for 6G Wireless Systems This research explores mathematical modeling for advanced signal processing, super diversity, holographic MIMO, and large-scale antenna arrays, while addressing hardware constraints and energy-efficient processing. This research is relevant to industry, driving innovation in next-generation wireless systems and shaping future communication networks. Compatible Tracks: TLC, ELN, CSE Number of positions: 1

Proposers: Mattia Brambilla, Monica Nicoli Topic: Communication and processing for smart factory and connected mobility The context is of real-time cooperative systems for smart road (connected vehicles) or smart factory (connected industrial assets/robots). Design and testing of experimental solutions for cooperative localization, sensing and mapping. Experimental activities are included with the IoTLab team in Bovisa Campus. Compatible Tracks: TLC Positions available: 2

Proposers: Matteo Cesana Topic: Modelling the Edge-Cloud Continuum In 5G, multi-access edge computing enables the applications to be offloaded to near-end edge servers for faster response. This research targets queuing theory-/AI-based models to assess different deployment strategies for applications in the edge-cloud continuum. Compatible Tracks: TLC, CSE, ATM Positions available: 1

Research topics in Bioengineering

Proposers: Alberto Antonietti Topic: Modelling neural plasticity with realistic biological models Neural plasticity is a complex and fascinating topic, to unveil its mysteries computational models of neural plasticity can imitate learning behavior in neural systems. The project goal is to develop biologically detailed models of plasticity to study how learning and memory can happen.