Error-correcting codes for IP/UDP streams with application to motorsport competitions.Contact person
LUCA BARLETTAEmail:
luca.barletta@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Dr. Antonino FavanoDescription
Description:
Master Thesis in collaboration with Marelli Motorsport
The thesis will focus on the analysis, design, and evaluation of advanced error correction techniques to improve the reliability and efficiency of wireless telemetry, audio, and video communications in motorsport competitions. Specifically, forward error correction algorithms will be investigated to complement the conventional protection mechanisms provided by existing radio technologies and enable more robust transmission from high-speed mobile terminals, installed on racing vehicles, to trackside network infrastructures. The proposed solutions will target limited redundancy and computational complexity, low latency, and efficient operation on resource-constrained hardware, while ensuring reliable and transparent integration with existing communication systems. Different coding strategies and their optimization according to channel conditions and traffic requirements will be evaluated, with the objective of improving end-to-end reliability in challenging high-mobility wireless environments.
Your tasks:
- Literature review of channel coding and error correction algorithms.
- Development of a data-driven channel model based on real-world measurements.
- System requirements analysis and selection of algorithms aligned with the identified requirements.
- Testing and validation of the selected algorithms in a simulation environment, preferably Linux-based.
- (Optional) Experimental validation on the target hardware through laboratory/bench tests.
- (Optional) Field validation during a motorsport event, comparing system performance with and without the selected algorithm.
Your profile:
- Familiarity with channel coding and error correction techniques, as well as wireless communication systems.
- Programming skills in C/C++, Python, MATLAB, or similar technical computing environments.
- Experience with Linux-based development environments is recommended, but not mandatory.
Hide Description
Microphysiological device for studying intestinal permeabilityContact person
MONICA SONCINIEmail:
monica.soncini@polimi.itStudy course: Biomedical Engineering
Other members of the research group:
Alessandro MarchesiniWeb page:
https://www.biomech.polimi.it/?page_id=1267Description
Description:
Development of a microphysiological device for studying intestinal permeability
With a mucosal surface area of approximately 32 m², the intestine represents one of the major interfaces between the human body and the external environment. Diet, obesity and alterations of the gut microbiota can impair intestinal barrier integrity, promoting the translocation of microbial products and metabolites into the circulation and potentially contributing to systemic inflammation. This phenomenon is particularly relevant considering the increasing prevalence of obesity, which currently affects approximately one in eight people worldwide. High-fat diets can alter intestinal tight junctions, mucus and microbiota, ultimately increasing barrier permeability.
A relevant example is MASLD (Metabolic dysfunction-Associated Steatotic Liver Disease), which affects approximately 30% of the global adult population and can progress, in advanced cases, toward fibrosis, cirrhosis and hepatocellular carcinoma. Alterations of the gut–liver axis and increased translocation of microbial products such as lipopolysaccharide (LPS) are among the mechanisms investigated in disease progression. Altered intestinal permeability is also being studied in cardiovascular diseases and, through the gut–brain axis, in neurodegenerative disorders such as Alzheimer's and Parkinson's disease, as well as in the interplay between gut microbiota, obesity and breast cancer.
A modular microphysiological platform is currently being developed at Politecnico di Milano to reproduce these interactions in vitro. The thesis will focus on the design and development of an automated device for the intestinal module, integrating perfusion, controlled mechanical stimulation and sensor-based permeability monitoring within a single system.
The project will cover the entire device development process, including system architecture definition, CAD design, rapid prototyping through laser cutting and 3D printing, design of components for injection molding, PCB development and testing, integration of sensors, actuators and electronics, and programming of the control system. The final prototype will be characterized and optimized for integration into the microphysiological platform.
The activities will be carried out at the ATTiC Lab and µBS Lab of Politecnico di Milano.
Starting date: March 2027
Hide Description
Automated Platform for Pancreatic Tissue Engineering ApplicationsContact person
MONICA SONCINIEmail:
monica.soncini@polimi.itStudy course: Biomedical Engineering
Other members of the research group:
Elia PederzaniWeb page:
https://www.biomech.polimi.it/?page_id=1267Description
Description:
Type 1 diabetes requires the administration of exogenous insulin, which, however, cannot fully reproduce the dynamic regulation of blood glucose provided by the endocrine pancreas. Pancreas or pancreatic islet transplantation represents a potential replacement strategy, but its clinical application is strongly limited by donor availability, the need for immunosuppression, and challenges related to graft engraftment and vascularization. In this context, the development of a vascularized and functional bioartificial pancreas through tissue engineering approaches represents a promising strategy for restoring pancreatic endocrine function. Recent studies have shown that decellularized extracellular matrix scaffolds, appropriately repopulated with endocrine and endothelial cells, can support the organization, maturation, and functionality of the engineered construct.
The Master's thesis project will focus on the experimental development of an automated platform for the recellularization and dynamic culture of a decellularized organ. The platform will be designed to enable the controlled seeding of different cell populations through specific anatomical access points, followed by fluid-induced physicochemical stimulation of the construct through perfusion. The aim is to establish reproducible culture conditions that support the formation of functional tissue.
The main activities will include a feasibility study and definition of the platform architecture, CAD design of its components, fabrication of a functional prototype, and its characterization through fluid-dynamic and functional testing. Particular attention will be devoted to the automation of cell-seeding and perfusion procedures and to the control of the main process parameters.
The activities related to platform development will be carried out at the laboratories of Politecnico di Milano, in collaboration with the Diabetes Research Institute at San Raffaele Hospital.
Start date: October 2026 / March 2027
Hide Description
Bioreactor to study immune recruitment in breast cancerContact person
MONICA SONCINIEmail:
monica.soncini@polimi.itStudy course: Biomedical Engineering
Other members of the research group:
Alessandra Rando, Alessandro MarchesiniWeb page:
https://www.biomech.polimi.it/?page_id=1267Description
Description:
Triple-negative breast cancer is an aggressive form of breast cancer characterized by complex interactions among tumor cells, stroma, vasculature, and the immune compartment. Investigating immune cell recruitment within the tumor microenvironment is therefore relevant to better understand the mechanisms underlying disease progression.
This thesis aims to develop a multicellular in vitro model consisting of stromal fibroblasts, triple-negative breast cancer spheroids, and vascular endothelium, integrated into a fluidic bioreactor. The system will be designed to enable the recirculation of culture medium containing macrophages, in order to investigate their recruitment and interaction with the tumor microenvironment. The experimental work will include the development and optimization of the bioreactor and fluidic conditions (CAD design, laser cutting, 3D printing...), as well as the establishment and characterization of the biological model.
Activities related to bioreactor development will be carried out at the ATTiC Lab of Politecnico di Milano, while the development and characterization of the in vitro model will be performed at Ospedale San Raffaele.
Starting date: October 2026/March 2027
Hide Description
Tricompartmental system to study the metastatic potential of colorectal cancerContact person
MONICA SONCINIEmail:
monica.soncini@polimi.itStudy course: Biomedical Engineering
Other members of the research group:
Alessandra RandoWeb page:
https://www.biomech.polimi.it/?page_id=1267Description
Description:
Colorectal cancer is one of the leading causes of cancer-related mortality worldwide, mainly due to the development of metastases. Its metastatic potential can be investigated by assessing the ability of cancer cells to extravasate, migrate, and colonize new biological sites. Current in vitro models used to study these processes are often compartmentalized systems in which the different compartments are separated by artificial membranes (polycarbonate, polyethylene...).
This thesis aims to develop a membrane-free tricompartmental in vitro model composed of cells and extracellular matrix (ECM), reproducing an endothelial–stromal–epithelial organization to study colorectal cancer cell extravasation and migration without introducing artificial barriers that could hinder the metastatic process recapitulation.
The experimental work will include optimization of a membrane-free device prototype (CAD design, laser cutting, 3D printing, silicone mold fabrication…) together with the development of a multicellular and multicompartmental in vitro model including intestinal epithelium, vascular endothelium, stromal fibroblasts, and colorectal cancer cells.
Experimental activities will be carried out at the ATTiC Lab of Politecnico di Milano.
Starting date: October 2026/March 2027
Hide Description
Scaling Analysis of Certified Maximum-Likelihood Decoding in Quantum LDPC Codes Across Syndrome Extraction RoundsContact person
LUCA BARLETTAEmail:
luca.barletta@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Mr. Michele BanfiDescription
Description:
The thesis will focus on the numerical and algorithmic evaluation of Maximum Likelihood (ML) decoding for quantum low-density parity-check (QLDPC) codes, specifically investigating the impact of the amount of measurements N_c. While fault-tolerant quantum memory protocols require multiple rounds of measurements to achieve fault tolerance, scaling N_c significantly increases the complexity for decoding algorithms. Focusing on representative small QLDPC codes, the candidate will analyze how the ML error rate evolve as N_c change.
Your tasks:
- Familiarize with quantum error correction and different noise models.
- Study the framework used to calculate ML bounds
- Set up Monte Carlo simulation sweeps for selected small QLDPC codes over varying numbers of measurements N_c.
- Evaluate the scaling behaviour of ML upper and lower bounds, tracking key metrics such as the fraction of unresolved simulation episodes and computational budgets (of the ML framework) as a function of N_c.
- Benchmark practical decoders against certified ML bounds across the sweep of N_c rounds to evaluate whether the optimality gap widens or stabilizes over extended execution depths.
Your profile:
- Background in coding theory and/or quantum computing.
- Proficiency in Python for scientific simulations and data analysis.
- Ability to work independently and interest in fault-tolerant quantum computing architectures.
- Good knowledge of the English language.
Hide Description
Mesh Networking for Drone SwarmsContact person
LUCA BARLETTAEmail:
luca.barletta@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Description
Description:
Master thesis in collaboration with Prof. Stefano Rini (DLR, Munich, Germany). Possibility to spend a period abroad at DLR, Munich, Germany.
The thesis will focus on the analysis and evaluation of advanced 5G networking protocols to support efficient and reliable communications in drone swarms, considering also the application of ML/RL solutions. Specifically, the use of DECT NR, world’s first non-cellular radio standard to be formally approved as part of the 5G standards by the ITU, will be investigated. The solution enables decentralized protocols in license-exempt spectrum, and can be a fundamental enabler for advanced drone communications.
Your tasks:
- Familiarize with the physical and MAC layer of the DECT-NR+ standard
- Study the application of the solution to drone communications, considering traffic requirements as well as constraints in terms of topology and channel impairments
- Evaluate by means of simulations, the protocol operations and evaluate its suitability for mesh networking in drone swarms for selected applications.
- Consider the use of machine learning/reinforcement learning solutions to design semantic networking protocols for drone swarms that communicate using DECT NR+
Your profile:
- Good knowledge of communication systems and signal processing, with particular emphasis on PHY and MAC level protocols.
- Knowledge of ML/RL algorithms
- Programming skills (C/C++) and willingness to learn new tools
- Previous experience with event-driven network simulations is a plus
- Ability to work independently and interest in interdisciplinary topics
- Good knowledge of English language
Hide Description
Semantic-aware PHY-Layer Techniques for Cooperative Drone Swarm CommunicationContact person
LUCA BARLETTAEmail:
luca.barletta@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Description
Description:
Master thesis in collaboration with Prof. Stefano Rini (DLR, Munich, Germany). Possibility to spend a period abroad at DLR, Munich, Germany.
The thesis focuses on the design and evaluation of physical layer (PHY) techniques for UAV swarm networks. While standard communications focus on bit-error-rate, the project explores a goal-oriented approach: extracting only the information that is actually relevant for swarm operations (e.g., for localization, navigation, communication) directly from the radio signal. This semantic, data-reduced representation can significantly cut the communication overhead needed for continuous swarm coordination. The work complements higher-layer protocol development (e.g., MAC/DECT NR+) by defining the PHY-layer constraints and data formats required for reliable cooperative operation.
Your tasks:
- Study how to extract relevant, task-specific information (i.e., features) from wireless signals, rather than transmitting raw data, to reduce payload size while preserving the semantic information required for cooperative localization and tracking
- Evaluate the approach through software simulations, analyzing the impact of channel effects (e.g., multipath, Doppler) on the accuracy of the extracted information necessary for the upper layers to operate effectively
- Gain hands-on experience testing your methods on real hardware platforms (e.g., SDR, UWB modules, flying drones), implementing the signal processing pipeline for real-time operation in airborne environments
- Collaborate with the team working on higher-layer protocols to define the best trade-off between raw and processed data transmission to maximize swarm efficiency
Your profile:
- Solid background in communication systems, signal processing, and wireless networks
- Interest in AI-based signal processing (feature extraction, classification) and semantic communications
- Experience with SDR platforms, radio hardware, and embedded systems (testing software on hardware) is a plus
- Programming skills in C/C++, MATLAB, and Python
- Ability to work independently and interest in interdisciplinary research topics (PHY protocols, AI, and hardware)
Hide Description
Performance modeling of HLS-generated acceleratorsContact person
SERENA CURZELEmail:
serena.curzel@polimi.itStudy course: Computer Science and Engineering
Description
Description:
Bambu HLS generates a hardware accelerator along with a C model used to obtain an (almost) accurate account of the number of clock cycles without running RTL simulation. The goal of the thesis is to improve the model by adding support for modeling external memory accesses so that an explanation can be provided to the user as to which clock cycles are used for calculation and which ones are spent waiting for memory.
Hide Description
In between Computer Vision and Computer Graphics: Processing Deformable Shapes with TransformersContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
Description
Description:
3D shapes (like a 3D model of a human or an animal) are central to many scientific and practical problems, ranging from Computer Graphics (e.g., video games) and Robotics to Medical Imaging and 3D Computer Vision. In this context, a relevant problem is shape matching, namely finding correspondences between two shapes. This is highly challenging as shapes can be deformable and even referring to different semantic subjects (e.g., matching the head of a cow with that of a tiger). While the case of two shapes has been satisfactorily resolved, other scenarios remain open, like the case of partial shapes (where part of the subject is not available) or multiple (more than two) shapes. This thesis will explore Transformer-like architectures to solve these relevant open problems.
References:
[1] https://arxiv.org/pdf/2411.03511
[2] https://onlinelibrary.wiley.com/doi/full/10.1111/cgf.70525
[3] https://arxiv.org/pdf/2409.13291
Hide Description
Neural Audio Codec Disentanglement for Synthetic Speech ForensicsContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Mirco Pezzoli, Viola NegroniDescription
Description:
Synthetic speech signals contain multiple intertwined factors, including linguistic content, speaker identity, prosody, recording conditions, and traces introduced by the generative model itself. Neural audio codecs provide structured latent representations of speech and may offer a way to separate these factors and isolate information specifically associated with the speech generation process.
The goal of this thesis is to investigate whether neural audio codec representations can be exploited to disentangle generator-specific traces from the other components of a speech signal. A first objective will be to determine whether such representations enable reliable attribution of synthetic speech to the model or generation pipeline used to produce it.
The thesis will then investigate whether generator-related information can be selectively manipulated while preserving the perceptual content and quality of the speech signal. This will enable a controlled study of how synthetic speech detectors respond to generator-specific traces, including whether modifying or suppressing such traces can cause attribution or detection systems to fail.
The final objective is to better understand which generator-specific characteristics are encoded in synthetic speech and how they can be exploited for attribution, detector interpretation, and robustness evaluation.
Hide Description
Acoustic Environment Inference from Processed Audio CallsContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Mirco Pezzoli, Federico Miotello, Fabio AntonacciDescription
Description:
A microphone recording contains acoustic cues related to the environment in which it was captured, including reverberation, background noise, and other characteristics of the recording space. These cues can potentially be exploited to estimate properties of the acoustic environment. However, modern communication platforms such as Microsoft Teams, Webex, and similar services apply extensive audio processing, including denoising, dereverberation, automatic gain control, and speech enhancement, which may significantly alter or remove such information.
The goal of this thesis is to investigate how much information about the original acoustic environment survives the processing chain of modern audio communication platforms and whether it can be recovered from the audio available at the receiving client.
The work will first study acoustic environment estimation from clean microphone recordings and then analyze how relevant cues and estimated parameters change after transmission through different communication platforms and processing configurations. The final objective is to assess which properties of the original recording environment, if any, can still be reliably inferred from the received audio and to develop methods for their estimation.
Hide Description
Knowing When Not to Decide: Opt-Out Strategies for Synthetic Media DetectionContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Viola Negroni, Daniele Ugo LeonzioDescription
Description:
Synthetic media detectors are typically designed to always produce a prediction, even when the analyzed content significantly differs from the data encountered during training. In real-world forensic applications, however, a detector should also be able to recognize when a sample falls outside its domain of competence and therefore avoid producing an unreliable decision.
The goal of this thesis is to develop opt-out strategies for synthetic media detectors, enabling them to determine whether a given audio, image, or video can be reliably analyzed. The work will investigate how internal representations, prediction confidence, uncertainty estimates, or out-of-distribution detection techniques can be used to characterize the detector's domain of competence.
The final objective is to design a selective detection framework in which the system can either classify a sample as real or synthetic, or abstain from making a decision when the available evidence is insufficient or unreliable.
Hide Description
Agentic AI for Orchestrating Multimedia Forensic DetectorsContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Giovanni Affatato, Daniele MoroDescription
Description:
Modern multimedia forensics relies on a growing collection of specialized tools addressing different tasks, such as synthetic media detection, manipulation detection, source attribution, and metadata analysis. Selecting which tools to apply and interpreting their potentially conflicting outputs is itself a complex reasoning problem.
The goal of this thesis is to develop an agentic AI framework for orchestrating multiple forensic detectors. Each detector will be exposed through a specialized agent capable of operating a specific forensic tool and interpreting its output, while a higher-level orchestrator will dynamically select and combine the appropriate agents according to the analyzed content and the evidence collected during the investigation.
The thesis will investigate tool selection, multi-agent coordination, evidence aggregation, uncertainty handling, and the use of LLM-based reasoning to produce a coherent and interpretable forensic assessment.
Hide Description
Analyzing Embedding Shifts for Robust Synthetic Speech DetectionContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Viola Negroni, Daniele Ugo LeonzioDescription
Description:
Synthetic speech detectors typically rely on learned representations that capture discriminative characteristics of real and generated audio. However, common editing and post-processing operations may alter these representations and significantly affect detector performance.
The goal of this thesis is to study how speech embeddings evolve when audio signals undergo operations such as compression, resampling, filtering, noise addition, or other forms of editing. The analysis will investigate the trajectories induced by these transformations in the embedding space and their relationship with the decisions made by synthetic speech detectors.
A further objective is to explore whether the observed embedding shifts can be exploited at test time, for example through controlled transformations of the input signal, to obtain more robust or reliable synthetic speech detection.
Hide Description
LLM-Assisted Image Source Attribution through Metadata AnalysisContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Daniele MoroDescription
Description:
Image files contain rich metadata that may provide useful forensic information about their origin and processing history. However, interpreting metadata is challenging because different cameras, smartphones, software tools, and online platforms generate heterogeneous combinations of fields and values.
The goal of this thesis is to develop an intelligent metadata analysis system for image source attribution. The system will combine a structured and indexed database of metadata collected from known sources with Large Language Models (LLMs) capable of reasoning over the metadata extracted from an unknown image.
The thesis will investigate how retrieval and LLM-based reasoning can be combined to identify the most likely acquisition device, software, or processing pipeline, while also providing interpretable evidence supporting the attribution decision.
Hide Description
Calibration and Fusion of Synthetic Media DetectorsContact person
PAOLO BESTAGINIEmail:
paolo.bestagini@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Viola NegroniDescription
Description:
Synthetic media detection is increasingly addressed through specialized detectors targeting different manipulation techniques, or generative models. However, the scores produced by different detectors are often not directly comparable and may exhibit significantly different calibration properties.
The goal of this thesis is to investigate score calibration and fusion strategies for synthetic media detection, considering audio or images. The work will explore how the outputs of heterogeneous detectors can be transformed into reliable and comparable confidence estimates and subsequently combined to improve detection robustness and generalization.
Possible directions include classical calibration and score-level fusion techniques, learning-based approaches, uncertainty estimation, and strategies for combining detectors when their reliability varies across different types of content or manipulation.
Hide Description
Fault detection in HLS-generated acceleratorsContact person
SERENA CURZELEmail:
serena.curzel@polimi.itStudy course: Computer Science and Engineering
Description
Description:
This thesis will investigate the integration of compiler-based fault-detection techniques into an open-source HLS flow, extending LLVM-level protection against single-event upsets to hardware accelerators. The work will evaluate how different protection strategies affect fault-detection coverage, performance, area, and power, and will explore HLS-specific optimizations to reduce their hardware overhead.
Hide Description
Exploration of custom floating-point acceleratorsContact person
SERENA CURZELEmail:
serena.curzel@polimi.itStudy course: Computer Science and Engineering
Description
Description:
The Bambu HLS tool supports the generation of custom floating-point processing units, e.g., featuring an arbitrary number of bits for mantissa and exponent. It is however difficult for the designer to decide which configuration is most suitable under performance, area, and accuracy requirements. This thesis will investigate automated design space exploration tools (possibly also ML/AI) that can assist this decision.
Hide Description
A Syntactic Algebra for Operator Precedence LanguagesContact person
MATTEO PRADELLAEmail:
matteo.pradella@polimi.itStudy course: Computer Science and Engineering
Description
Description:
Operator Precedence Languages form a large subclass of deterministic context-free languages that preserves several properties of regular languages, including Boolean closure, decidable inclusion, and logical characterizations. The aim of this thesis is to investigate an algebraic characterization of OPLs analogous to the syntactic monoid of regular languages.
Hide Description
Model Checking of Operator Precedence LanguagesContact person
MATTEO PRADELLAEmail:
matteo.pradella@polimi.itStudy course: Computer Science and Engineering
Description
Description:
Model checking is an established technique for automatically verifying whether a system satisfies temporal properties. Traditional approaches based on finite-state models, however, are not sufficient to represent the potentially unbounded call stack of recursive programs. Operator Precedence Languages (OPLs) provide a suitable formal model for such systems. They are more expressive than Visibly Pushdown Languages and can naturally represent complex control-flow mechanisms, including recursive procedure calls, exception handling, and stack unwinding.
This thesis will investigate model-checking techniques for programs represented by Operator Precedence Automata. Properties of program executions will be specified using Precedence Oriented Temporal Logic (POTL), which extends classical temporal logic with operators that navigate the hierarchical structure induced by operator-precedence relations. The starting point will be POMC, an existing model checker that translates POTL formulas into Operator Precedence Automata or Büchi Automata and verifies both finite and infinite program executions.
Hide Description
Operator-Precedence Languages for the Parallel and On-the-Fly Analysis of Big DataContact person
MATTEO PRADELLAEmail:
matteo.pradella@polimi.itStudy course: Computer Science and Engineering
Description
Description:
The thesis will investigate the use of Operator-Precedence Languages as a formal foundation for the efficient analysis of large structured datasets represented in formats such as XML and JSON.
The main idea is to exploit the local parsability property of Operator-Precedence Languages. Unlike general context-free languages, their syntactic structure can be identified using limited contextual information. This property may allow a large document to be divided into several fragments that can be parsed and analyzed independently and in parallel. Each fragment would be associated with a compact summary describing both its internal structure and the syntactic information required to combine it with adjacent fragments.
The research will study how such summaries can be formally defined and composed, so that the result of analyzing two consecutive fragments is equivalent to analyzing their concatenation directly. This approach could support parallel parsing on multicore or distributed architectures, as well as streaming and on-the-fly analysis, where properties are checked as soon as the relevant portion of the input becomes available.
Hide Description
Middleboxes for network segmentationContact person
GIACOMO VERTICALEEmail:
giacomo.verticale@polimi.itStudy course: Computer Science and Engineering, Telecommunications Engineering
Description
Description:
Modern industrial control systems (Operational Technology, OT) rely on strict network segmentation to enforce security boundaries between subsystems. Communications across security zones are typically restricted and must traverse dedicated middleboxes that inspect and validate exchanged information before forwarding it. Similar requirements are emerging in new application domains, including the interaction of AI agents operating across different security domains.
A related challenge arises in the context of the 5G Service-Based Architecture (SBA), where multiple logical networks (network slices) coexist on a shared physical infrastructure while maintaining distinct security and performance requirements. In such environments, controlled inter-domain communication mechanisms are required to enforce isolation without compromising operational efficiency.
Our recent research has proposed two alternative approaches for implementing secure middleboxes. The first relies on advanced cryptographic techniques that allow the middlebox to perform security checks while preserving the confidentiality of the exchanged data [1]. The second exploits credential delegation mechanisms to authorize cross-domain operations with significantly lower computational overhead, at the cost of weaker security guarantees [2].
The objective of this thesis is to deploy and experimentally evaluate one or both of these approaches in realistic OT and/or 5G SBA scenarios. The work will include:
- Deployment and configuration of a secure middlebox prototype;
- Integration within representative segmented-network architectures;
- Experimental performance evaluation, including latency, throughput, scalability, and resource consumption;
- Analysis of the security-performance trade-offs offered by cryptographic processing and credential-delegation techniques;
- Assessment of the suitability of the proposed solutions for emerging use cases, including communication between AI agents operating across security zones.
The expected outcome is a quantitative comparison of secure middlebox architectures and a characterization of their applicability to industrial and 5G environments with stringent security and performance requirements.
**References**
[1] https://ieeexplore.ieee.org/abstract/document/10588930
[2] https://ieeexplore.ieee.org/abstract/document/11080543
Hide Description
From Networked Publics to Algorithmic Audiences: The Broadcastization of Social Media PlatformsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Web page:
https://pierri.faculty.polimi.itDescription
Description:
This thesis investigates how and to what extent social media platforms are transforming into algorithmically governed broadcast platforms. The project examines the shift from social-graph-based interaction and user-generated participation toward interest-based recommendation, passive viewing, creator-centric visibility, and AI-mediated content production.
The study will operationalize “broadcastization” through measurable indicators such as the prominence of recommendation feeds, the declining role of friend/follower networks, changes in posting versus viewing behavior, concentration of visibility among professionalized creators, and the presence of synthetic or AI-assisted content. A comparative analysis of platforms such as TikTok, Instagram Reels, YouTube Shorts, X/Twitter, and Facebook will combine interface analysis, platform documentation, content sampling, and engagement metrics.
The thesis asks how social media platforms are becoming broadcast platforms, what technical and economic mechanisms drive this transformation, and what consequences follow for user agency, participation, visibility, and information exposure. The expected contribution is a conceptual and empirical framework for studying the transition from networked publics to algorithmic audiences, clarifying the democratic and cultural stakes of post-social media environments.
Submission targets:
The ACM Web Conference, CHI, CSCW, ICWSM
Suggested reading:
Törnberg P, Rogers R. Towards a Post-Social Media Studies [Internet]. SocArXiv; 2026. Available from: osf.io/preprints/socarxiv/6nue7_v1
Hide Description
Agentic AI for Predictive and Cooperative V2X Communication in Urban Mobility ScenariosContact person
FRANCESCO LINSALATAEmail:
francesco.linsalata@polimi.itStudy course: Automation Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Roberto Pegurri, Antonio Capone, Maurizio MagariniDescription
Description:
Future Cooperative, Connected and Automated Mobility (CCAM) systems will operate in highly dynamic urban environments characterized by heterogeneous connectivity, partial observability, and stringent safety requirements. In these scenarios, communication, sensing, and coordination mechanisms cannot be designed independently, as the availability and quality of information directly affect cooperative decision-making and traffic safety.
This thesis investigates the use of Agentic Artificial Intelligence (Agentic AI) for predictive and cooperative V2X communication systems. The objective is to design distributed AI agents capable of perceiving the communication and environmental context, reasoning on network conditions and sensing information, and autonomously adapting communication and coordination strategies in real time. Particular attention will be devoted to predictive communication techniques exploiting digital-twin information, sensing-aware representations, and multi-frequency vehicular communication models.
The activity will focus on urban scenarios involving connected and non-connected vehicles, roadside infrastructure, and vulnerable road users under heterogeneous communication conditions. The student will investigate how AI-driven agents can support adaptive information dissemination, cooperative perception, task orchestration, and communication-aware coordination while accounting for latency, reliability, and observability constraints.
The work will combine simulation and experimental activities using advanced vehicular networking and digital-twin frameworks, including tools such as NS-3, SUMO, Sionna, and ray-tracing-based communication models. Depending on the thesis scope and student interests, the work may also include the integration of sensing data, distributed learning approaches, and experimental validation on real communication platforms.
The thesis is part of ongoing research activities at Politecnico di Milano with Institute of Science Tokyo formerly Tokyo Tech on next-generation V2X systems, digital twins, and AI-driven wireless communication for future CCAM and 6G vehicular networks.
Hide Description
Towards Safe Social Media for Adolescents: A Privacy-Preserving Mobile App Using On-Device Large Language ModelsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Web page:
https://pierri.faculty.polimi.itDescription
Description:
This thesis aims to design and develop a privacy-aware mobile application that uses on-device Large Language Models to monitor social media content consumed by teenagers and detect harmful or dangerous material, with a core focus on simulating realistic teenage accounts on Instagram and WhatsApp for safe, controlled evaluation.
The project will build an Android/iOS app that captures on-screen content and chat messages locally, and construct a simulation framework that generates synthetic teenage Instagram feeds, stories, reels, and direct messages as well as WhatsApp group and one-on-one chats, programmatically varying risk types (e.g., cyberbullying, grooming, self-harm), language styles, and media representations.
It will explore on-device LLM techniques such as privacy-preserving prompt engineering, lightweight adapter fine-tuning on simulated data, and representation steering to balance detection accuracy with computational efficiency.
Hide Description
Development of a multimodal hearing testing platform for assessment of human individual sensory, behavioral, and listening effort responsesContact person
ALESSIA PAGLIALONGAEmail:
alessia.paglialonga@polimi.itStudy course: Biomedical Engineering
Other members of the research group:
Riccardo BarbieriWeb page:
https://shorturl.at/GiSyJDescription
Description:
The aim of this thesis is to develop and test a multimodal hearing testing platform to assess and characterize human individual listening responses and listening effort. This study will recruit among a general population of adults (age>18) with and without hearing impairment and will use a multimodal testing procedure able to characterize participants with varying auditory profiles. The range of measures used includes self-assessment surveys, audiometric testing, new speech in noise tests, and physiological measures (using unobtrusive sensors) in various experimental setups. Advanced signal processing techniques and data analysis methods, including AI, will be used to analyze the recorded data, supporting the development of new, multidimensional models of hearing impairment and listening effort.
Hide Description
Adaptive Instruction Prefetching for Modern SoC WorkloadsContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
Modern SoC performance is increasingly limited by the "front-end bottleneck," where complex software (microservices, AI, large-scale apps) causes frequent instruction cache misses. This thesis investigates and develops novel instruction prefetching mechanisms designed for non-linear execution patterns. Using cycle-accurate simulators (e.g., Verilator) and then FPGA prototyping, the research will evaluate existing state-of-the-art algorithms against modern traces to propose a new, resource-efficient prefetcher that maximizes throughput while minimizing power and area overhead.
Hide Description
Radio-frequency electromagnetic compatibility analysis in small and micro satellites (Master's Degree Thesis)Contact person
XINGLONG WUEmail:
xinglong.wu@polimi.itStudy course: Electrical Engineering, Electronics Engineering, Telecommunications Engineering
Other members of the research group:
Ludovica Illiano and the industry supervisor at OHB ItaliaWeb page:
Details see: Bacheca Tesi (Thesis bulletin board) accessible through the POLIMI online servicesDescription
Description:
(Thesis-oriented internship at OHB Italia)
Small and micro satellites present a significant challenge for radio-frequency compatibility verification by analysis, a fundamental step that must be performed within the design loop long before verification by test, which usually occurs at a much later stage of the spacecraft production cycle. The compact size of these platforms results in very limited distances between antennas, typically far smaller than the separation required to remain within the validity limits of the analytical methodologies commonly used for larger spacecraft.
In this context, the aim of the thesis is to develop an analytical approach to address this challenge, specifically:
To use FEKO to develop antenna models typically employed on small and micro platforms (e.g., patch antennas).
To use the FEKO optimization tools to tune and fit these antenna models using information provided by the antenna suppliers.
To use the developed antenna models to derive coupling factors between different antennas at both in band and out of band frequencies in real case scenarios.
Hide Description
Development of Side-Channel Attacks on Commercial TPUs: Physical and Microarchitectural Threat Landscapes in Edge AIContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Prof. Alberto Bosio (University of Lyon, France)Web page:
https://cassano.faculty.polimi.it/Description
Description:
The proliferation of edge computing has pushed complex machine learning models onto low-power, specialized hardware such as the Google Coral Edge TPU and Axelera AI accelerators. While these devices offer high performance per watt, their physical deployment in uncontrolled environments exposes them to hardware-level attacks. This thesis investigates the susceptibility of commercial edge AI accelerators to physical side-channel analysis (SCA) and fault injection techniques. The research aims to quantify the risk of proprietary model extraction (IP theft) and adversarial inference disruption, ultimately proposing lightweight hardware and software countermeasures suitable for resource-constrained edge environments.
Hide Description
Comparative analyses of computational social science studies on Chinese social media platformsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Geng LiuDescription
Description:
This thesis aims to replicate and extend established findings from studies conducted on Twitter/X and Facebook to major Chinese social media platforms (e.g., Weibo, WeChat, Douyin). The project will examine whether patterns related to information diffusion, polarization, misinformation spread, coordinated behavior, or engagement dynamics hold across different platform architectures and regulatory contexts. Comparative analyses will assess similarities and divergences in network structures, content amplification mechanisms, and user interaction patterns. The goal of the thesis is to evaluate the generalizability of social media research findings across platforms and geopolitical contexts, contributing to a more globally grounded understanding of online information dynamics.
Submission targets: The ACM Web Conference (Oct-Nov)
Hide Description
Eat the Rich: Investigating public opinion on wealthy people after the United Healthcare CEO shootingContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Francesco CorsoDescription
Description:
This thesis aims to study the public opinion response after the fatal shooting of the United Healthcare CEO in early December 2024. This event has sparked a lot of controversy online, with the mainstream media taking a clear stance against the event, while a non-negligible amount of users online were praising and siding with the alleged shooter, Luigi Mangione.
This separation of the official mainstream narrative and the social reaction of the general public on social media is the object of our study.
We collected a dataset of news reporting the facts, together with reddit posts that mention these news. We have to extend this analysis with TikTok data, which will entail the use of video understanding and video summarization techniques.
Submission targets: The ACM Web Conference - ARR (Oct-Nov)
Hide Description
Probing conspiratorial mindset in Large Language ModelsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Francesco CorsoDescription
Description:
This thesis investigates whether and to what extent Large Language Models (LLMs) can exhibit or amplify a conspiratorial mindset, building on results presented in previous work. The project will operationalize key dimensions of conspiratorial thinking—such as intentionality attribution, distrust of institutions, and perception of hidden coordination—into measurable constructs. A benchmark dataset of targeted prompts across multiple domains (e.g., politics, health, climate) will be developed to elicit conspiratorial reasoning under different conditions (e.g., zero-shot prompting, instruction tuning). A dedicated evaluation pipeline combining automated metrics and human annotation will assess the prevalence and intensity of conspiratorial outputs. The goal is to produce a standardized benchmark and methodological framework to systematically evaluate downstream conspiratorial tendencies in LLMs and their implications for model safety and societal risk.
Submission targets: The ACM Web Conference (Oct-Nov), *CL (various deadlines: https://aclrollingreview.org/dates)
Reference: [2511.03699] Do Androids Dream of Unseen Puppeteers? Probing for a Conspiracy Mindset in Large Language Models
Hide Description
AI-enabled Federated Data Products for HealthcareContact person
PIERLUIGI PLEBANIEmail:
pierluigi.plebani@polimi.itStudy course: Computer Science and Engineering
Description
Description:
This thesis focuses on the design and implementation of federated data products in the healthcare domain, leveraging the principles of Data Mesh and aligning with the requirements of the European Health Data Space (EHDS). The central goal is to define engineering methods, architectures, and tools to enable a federated data ecosystem, where data remains distributed across organizational and national boundaries while being accessible in a secure, interoperable, and governed manner.
The work, relying on the adoption of Agentic AI paradigm, will address the challenges of integrating and managing heterogeneous and distributed healthcare data, including genomic data, medical imaging (e.g., DICOM), and clinical data (e.g., EHR, FHIR). A key focus will be on designing data product abstractions that can operate in a federated context, including data contracts, APIs, metadata, and service-level objectives, ensuring discoverability, trust, and reuse without centralizing data.
We propose the following thesis tracks:
- the design of federated and decentralized data architectures based on Data Mesh principles when considering clinical data
- the exploration of Agentic AI approaches (e.g., autonomous or semi-autonomous agents) to support tasks such as data product design, metadata enrichment, orchestration of data workflows, and intelligent data discovery in federated environments for clinical data
Hide Description
3D Reconstruction under Alternative Camera ModelsContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
Web page:
https://federica-arrigoni.github.io/Description
Description:
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 momument), which is observed from multiple cameras at different locations and viewpoints, generating the input images. The goal of this thesis is to develop novel 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).
Reference:
A taxonomy of structure from motion methods https://arxiv.org/abs/2505.15814
Hide Description
Metasurface-Enhanced Passive Patches for WiFi-Based Contactless Vital Sign MonitoringContact person
SILVIA MURAEmail:
silvia.mura@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electrical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Marouan MizmiziDescription
Data Science for SportsContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Description
Description:
In recent years, the proliferation of tracking data in soccer (particularly from GPS, optical tracking, and advanced scouting systems) has transformed the analytical landscape of the sport. While traditional metrics such as expected goals (xG) have played a pivotal role in bridging raw data and tactical insight, the growing availability of granular spatio-temporal information calls for a deeper, more nuanced understanding of goal-scoring dynamics.
This thesis aims to move beyond conventional xG frameworks by exploring where, when, and how goals are most likely to be scored. Specifically, it will investigate (1) which areas of the goal are statistically more likely to produce a successful shot, challenging the long-held assumption that certain “ideal” target zones (e.g., the top corners) are inherently superior, and (2) which on-field positions provide the most favorable conditions for scoring opportunities.
The analysis will leverage recent and high-resolution data from both the Men’s and Women’s top competitions, employing modern data science and machine learning techniques to uncover latent patterns in shot placement, build-up play, and positional dynamics. Beyond descriptive statistics, the research seeks to uncover predictive insights that can guide evidence-based decision-making in professional soccer.
The study will be carried out in collaboration with professional performance analysts, whose practical expertise will ensure the validity and applicability of the findings.
Hide Description
Sparse Training for Efficient On-Device LearningContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Hazem Hesham Yousef ShalbyDescription
Description:
Training neural networks directly on resource-constrained edge devices is increasingly desirable for privacy preservation, user personalization, and continual adaptation, yet it remains largely impractical due to the memory and computational demands of backpropagation. This thesis investigates sparse-to-sparse training, an approach that sparsifies the backward pass itself by selectively propagating and applying only a subset of gradient information at each update step, without altering the learning objective. We explore lightweight gradient selection strategies, including magnitude thresholding and top-k filtering, alongside structured sparsity patterns such as block- and channel-level updates that align with the computational primitives of embedded hardware. To ensure stable convergence under high sparsity, we study simple stabilization mechanisms including gradual sparsity scheduling and inexpensive gradient rescaling. The proposed methods are evaluated against dense baselines on representative small-to-medium architectures, measuring classification accuracy, convergence behavior, peak memory footprint, and training cost via latency and energy proxies, with direct on-device profiling where feasible. The expected outcome is a principled and practical recipe for on-device training, accompanied by concrete guidelines for selecting backward sparsity levels that balance computational efficiency with learning quality.
Hide Description
LLM reasoning and Agentic AIContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Description
Description:
Large Language Models can generate fluent text, but can they actually reason and act reliably? This thesis explores the frontier between language modeling and autonomous AI agents. The student will investigate how inference-time strategies, such as chain-of-thought prompting, self-verification, and structured search can improve multi-step reasoning in LLMs, and/or how these capabilities transfer to agentic settings where the model must plan, use tools, and execute actions over extended horizons. A central challenge is error compounding: small mistakes cascade across action chains, leading to task failure. The thesis will design and evaluate mechanisms for self-monitoring and failure recovery in LLM reasoning and/or LLM-based agents, with experiments on reasoning benchmarks and agentic tasks (e.g., code generation, web navigation, tool use). All experiments will be conducted using small and efficient open-weight language models, investigating how far lightweight models can be pushed in (agentic) reasoning tasks. The work sits at the intersection of natural language processing, planning, and AI safety.
Hide Description
Enhancing Autonomous UAV-Based Localization through Hybrid Signal Processing and Adaptive Mission PlanningContact person
FRANCESCO LINSALATAEmail:
francesco.linsalata@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Maurizio MagariniDescription
Description:
This thesis explores the use of autonomous unmanned aerial vehicles (UAVs) as passive sensing platforms for multi-user identification and localization in 5G cellular networks. Building on recent advances in UAV-based signal intelligence, the work aims to enhance the accuracy, robustness, and efficiency of user localization by combining advanced signal processing techniques with adaptive UAV mission planning. The research will investigate methods for extracting spatial information from uplink reference signals, addressing challenges such as multi-user interference, multipath propagation, and limited sensing time. In addition, the thesis will study intelligent trajectory optimization strategies that allow the UAV to dynamically adapt its flight path based on real-time signal measurements, balancing localization performance and energy consumption. The expected outcome is a scalable and infrastructure-independent framework for reliable user localization, with strong potential impact in emergency response, disaster recovery, and next-generation wireless network monitoring.
Hide Description
A new deep-learning paradigm for dense 3D ReconstructionContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
Web page:
https://federica-arrigoni.github.io/Description
Description:
This thesis explores the recent work entitled “VGGT: Visual Geometry Grounded Transformer”, which provides a new deep-learning paradigm for dense 3D reconstruction from arbitrary image collections. Specifically, the goal is to analyze limitations of VGGT and propose possible improvements.
Reference: https://vgg-t.github.io/
Hide Description
Good graphs versus bad graphs in 3D reconstruction from multiple imagesContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
Web page:
https://federica-arrigoni.github.io/Description
Description:
A powerful tool to study and solve relevant problems in Computer Vision (and, more precisely, in the sub-field of multi-view geometry) is representing cameras/images and their pairwise relations as vertices and edges of a graph. In this context, a prominent problem is 3D reconstruction from images and a relevant question is the following: which graphs are good and which graphs are bad for 3D reconstruction? This thesis will develop novel methods aimed at answering this question.
Reference:
F. Arrigoni, A. Fusiello, T. Pajdla. A direct approach to viewing graph solvability. ECCV 2024
https://www.ecva.net/papers/eccv_2024/papers_ECCV/html/126_ECCV_2024_paper.php
Hide Description
Reproducing Science with Generative AIContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Web page:
https://pierri.faculty.polimi.itDescription
Description:
This thesis explores the use of Large Language Models (LLMs) to support and automate the replication of scientific studies, with a particular focus on reproducing data analysis pipelines from published research. Replicability is a cornerstone of scientific integrity, yet many studies remain difficult to reproduce due to missing, incomplete, or poorly documented code. This work investigates whether LLMs can bridge this gap by reconstructing analysis workflows directly from research papers, leveraging natural language descriptions, figures, and available datasets.
The thesis will study scenarios both where original analysis code is available and where it is absent but the data is accessible. Methods will include prompting and agent-based approaches for pipeline reconstruction, code generation, parameter inference, and validation of reproduced results against reported findings. The evaluation will assess accuracy, robustness, transparency, and limitations of LLM-assisted replication, as well as the implications for open science, peer review, and automated reproducibility checks. Ultimately, the thesis aims to clarify the practical role LLMs can play in scaling scientific replication and improving research reliability.
Hide Description
Analysis of the Impact of Microarchitectural and Architectural Features of Modern Processors on the Susceptibility to Transient Execution AttacksContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
Thanks to the open-source RISC-V Instruction Set Architecture, this thesis aims at systematically analyzing how specific microarchitectural and architectural configurations of modern processors, including pipeline depth, cache timing, branch prediction algorithms, and execution latencies, impact on vulnerability to Transient Execution Attacks (TEAs). By utilizing a configurable RISC-V simulation/emulation environment, the thesis isolates these hardware parameters to measure their direct correlation with side-channel leakage rates and the success probability of speculative attacks like Spectre. The thesis aims to quantify how expanding the speculative window (via delayed resolution) or altering cache responses modifies the attack surface. Ultimately, this work will provide design guidelines for hardening RISC-V cores against transient execution vulnerabilities by identifying the most critical configuration weaknesses.
Hide Description
Development of Machine Learning Methodologies for Security Attacks Detection and Prevention Based on On-Chip MeasurementsContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis focuses on developing a machine learning (ML) system for detecting security attacks, such as HW Trojan horses and Transient Execution Attacks, by analyzing various system measurements, including microarchitectural data. The research will explore which system metrics are most effective for accurate detection and how the ML system can interact directly with hardware to ensure detection and reaction against attacks.
Hide Description
Studying Mechanisms of AI Search EnginesContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Web page:
https://pierri.faculty.polimi.itDescription
Description:
This thesis explores how AI search engines retrieve and generate information. Two directions are available:
Bias & Harmfulness – Auditing AI search outputs for bias, misinformation, and harmful patterns, and evaluating strategies to reduce these risks.
Generative Engine Optimization (GEO) – Investigating how content properties influence ranking in AI-generated search results and testing methods for optimizing visibility.
Both tracks involve building evaluation pipelines, running targeted experiments, and outlining guidelines for more transparent and reliable AI search systems.
Hide Description
Benchmarking cultural awareness in Large Language ModelsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Anna BernasconiWeb page:
https://pierri.faculty.polimi.itDescription
Description:
This thesis aims to design and evaluate methods for building cultural awareness into Large Language Models by operationalizing Erin Meyer’s Culture Map dimensions.
The project will construct a multimodal and multilingual dataset annotated along Culture Map variables and explore techniques such as prompt engineering, representation steering, and fine-tuning with culturally grounded exemplars. A dedicated evaluation pipeline will benchmark improvements in cultural sensitivity and communication appropriateness through ablation studies (dimension-specific, combined-dimensions, and culturally adaptive models).
Hide Description
Social Robots and SSHContact person
MARIAGRAZIA FUGINIEmail:
mariagrazia.fugini@polimi.itStudy course: Computer Science and Engineering
Description
Safety and Resilience in Smart EnvironmentsContact person
MARIAGRAZIA FUGINIEmail:
mariagrazia.fugini@polimi.itStudy course: Computer Science and Engineering
Description
Fraud Detection and AIContact person
MARIAGRAZIA FUGINIEmail:
mariagrazia.fugini@polimi.itStudy course: Computer Science and Engineering
Description
Designing DL applications resilient against hardware failuresContact person
CRISTIANA BOLCHINIEmail:
cristiana.bolchini@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Antonio MieleDescription
Description:
In recent years, a lot of effort has been devoted to estimating and evaluating the resilience of Deep Learning-based applications against the occurrence of faults in the underlying hardware, also trying to investigate the relative relevance of the network neurons and layers. The thesis wants to investigate methods and tools either to estimate the resilience of the application or to harden it so that even if faults affect the underlying hardware do not propagate to the DL outcome.
Hide Description
Development of Transient Execution Attacks against NIST Post-Quantum CryptographyContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis investigates the vulnerability of new NIST-standardized post-quantum cryptography (PQC) schemes, such as ML-KEM and ML-DSA, to timing-based side-channel attacks. The project's primary focus is on transient execution attacks (like Spectre), which can create subtle but critical timing leaks that are often missed in standard evaluations.
The first phase of the project involves building a high-precision tool to measure execution time. This tool will be used, along with statistical tests, to find and characterize any timing variations that depend on secret keys. Once a vulnerability is identified, a central objective is to develop a proof-of-concept Spectre-style attack. This attack will aim to demonstrate a practical method for recovering secret key material by exploiting a cache-timing channel.
Finally, the research will measure the effectiveness of different defenses. This analysis will quantify the impact of constant-time coding, various compiler settings, and specific CPU features. The results will establish how feasible these attacks are in practice, and the key deliverable will be a set of concrete recommendations for securing PQC in real-world applications.
Hide Description
Multicast-Capable NoC Routers for Coherence Traffic in RISC-V Multicore Clusters: Deadlock-Safe Routing and Protocol Co-DesignContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis develops hardware multicast/replication in NoC routers to accelerate coherence actions—invalidations, updates, and line delivery—to multiple sharer clusters in RISC-V–based systems. It co-designs multicast routing trees, VC allocation, and turn-model constraints to guarantee deadlock freedom while preserving QoS for latency-critical traffic. A cluster-aware RISC-V directory protocol selects unicast vs. multicast adaptively based on sharer sets and congestion. Benchmarks quantify reductions in NoC load, tail latency, and coherence transaction time versus unicast baselines.
Hide Description
Topology-Aware NUCA over NoC for RISC-V SoCs: Placement, Migration, and Coherence-Driven Data MovementContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis investigates a distributed last-level cache (NUCA) carved across NoC routers in RISC-V–based SoCs and co-optimizes line placement/migration with coherence behavior. Policies exploit network distance and sharer patterns to co-locate data near its “home” RISC-V cluster, with background migration throttled by NoC congestion. Coherence metadata and on-chip telemetry drive prefetching and replication of hot read-mostly lines. Results highlight access latency, traffic balance, and energy-delay product under mixed workloads.
Hide Description
Ultra-Compact NoC for RISC-V MCU-Class Multicores with Lightweight Coherence and Low-Power LinksContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
Targeting microcontroller-scale RISC-V multicores, this thesis proposes minimal-area routers and low-power links (SerDes, link-level flow control, clock/power gating) alongside a lightweight coherence protocol tailored to small clusters. It evaluates deterministic routing for timing predictability versus mildly adaptive schemes for congestion relief, proving deadlock/livelock freedom. A hierarchical “snoop-local, directory-global” design reduces broadcasts while keeping controller logic tiny. ASIC-style synthesis and FPGA emulation quantify area, timing, and energy vs. a shared bus.
Hide Description
Design and development of a custom RISC-V pipeline for two-way ranging (TWR) with ultra-wideband (UWB) technologyContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Luca MottolaWeb page:
https://cassano.faculty.polimi.it/Description
Description:
Goal of the thesis is to prototype a custom RISC-V pipeline to support two-way ranging (TWR) with ultra-wideband (UWB) technology. TWR is a distance measurement method that uses the round-trip time of a UWB signal exchanged between two devices to calculate the precise distance between them. TWR technology is used in mobile computing and robotics to implement location-aware functionality and robot coordination. Distance values are obtained from low-level features of UWB signals by applying a specific processing pipeline, which may be optimized with a dedicated RISC-V design.
Hide Description
Development of Novel Trusted Execution Environments (TEEs) (in collaboration with the European Space Agency)Contact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis focuses on the design and development of new Trusted Execution Environments (TEEs) to enhance the security of computing systems. The research will explore innovative TEE architectures that provide improved protection for sensitive data and operations, particularly in embedded systems, ensuring that security is maintained even in the presence of malicious attacks (Transient Execution Attacks like Spectre and Meltdown, for example) or software vulnerabilities.
Hide Description
Design of Integrated GPU Architectures for RISC-V System-on-Chip (SoC) Platforms (in collaboration with the European Space Agency)Contact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis explores the development of new integrated GPU architectures within RISC-V SoC platforms. By designing GPU modules that are seamlessly integrated with the RISC-V architecture, the research will focus on improving graphical processing efficiency and enabling more powerful and energy-efficient SoC solutions for diverse applications.
Hide Description
Exploiting GPU Microarchitectures for Novel Hardware Attacks: Investigating Memory Leaks and Arbitrary Code Execution Without Physical Access (in collaboration with the European Space Agency)Contact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis will investigate how modern Graphics Processing Units (GPUs) can be exploited to reveal sensitive data or run unauthorized code, all without requiring physical access or elevated permissions. By studying GPU-specific features, such as caching techniques, memory management, and parallel processing pipelines, the project aims to uncover new attack vectors and demonstrate them through proof-of-concept exploits. Finally, the research will suggest countermeasures and improved design strategies to enhance GPU security against these emerging threats.
Hide Description
Exploring Security Vulnerabilities in the OpenTitan Framework: A Comprehensive Analysis and Testing Approach (in collaboration with the European Space Agency)Contact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia Lazzeri, Gianluca Furano (European Space Agancy)Web page:
https://cassano.faculty.polimi.it/Description
Description:
This work focuses on an in-depth security evaluation of the OpenTitan framework which is an open-source silicon root of trust designed to reinforce trusted computing on hardware-level systems. The research will begin with a thorough review of OpenTitan’s architecture, security features, and implementation details to identify potential weakness points. Building on this understanding, the project will employ systematic testing methods, including static code analysis, fuzzing, and hardware penetration testing, to uncover vulnerabilities and assess the impact of any discovered flaws on system integrity.
Hide Description
