Sequenzializzazione degli ordini di produzione su macchine utensili nel MES MaDS: progettazione, sviluppo e analisi prestazionale data-drivenContact person
PIERLUIGI PLEBANIEmail:
pierluigi.plebani@polimi.itStudy course: Computer Science and Engineering
Description
Description:
MaDS è il MES di NEALIS, una piattaforma modulare pensata per connettere ERP, reparti e macchine, raccogliendo eventi di produzione in tempo reale. La tesi propone la progettazione e lo sviluppo di un modulo software che sia in grado di sequenzializzare gli ordini di produzione su macchine utensili. Dato un insieme di ordini (con attributi come quantità e tempi ciclo teorici), un obiettivo primario (rispetto della date di fine produzione) e vincoli operativi (compatibilità ordine-macchina, calendari di disponibilità, esecuzione concorrente di più ordini), il modulo assegna gli ordini alle risorse e genera una sequenza ottimizzata. L’approccio prevede modelli di scheduling vincolato e/o euristiche migliorative, con valutazione tramite KPI (ritardo, utilizzo macchina, throughput). In parallelo, viene sviluppato uno strumento di analisi che sfrutta i dati del MES per determinare i tempi ciclo reali per articolo e risorsa, confrontarli con i teorici e identificare le macchine più performanti. Il risultato è un ciclo chiuso pianificazione-misura-adattamento, a supporto di decisioni di produzione affidabili e data-driven.
Lo stack tecnologico utilizzato si basa su:
- Linguaggi di sviluppo PHP (Lavarel/Symfony) + javascript (Backbone.js, Bootstrap e Vue.js)
- Database relazionale PostgreSQL
- Database NoSQL MongoDB e Redis
- OS Linux
- Virtualizzazione docker
- Source management git (Bitbucket)
Il tema proposto è da svolgersi in collaborazione con Nealis (https://nealis.it)
Hide Description
AI-driven Policy Generation for Data Mesh Monitoring using OPAContact person
PIERLUIGI PLEBANIEmail:
pierluigi.plebani@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Valeria Maria FortinaWeb page:
https://plebani.faculty.polimi.it/Description
Description:
Data Mesh is an emerging paradigm for cross-organizational data sharing, emphasizing domain-oriented ownership, self-service data infrastructure, and decentralized governance. In multi-organization contexts, ensuring compliance with agreements on privacy, performance, visibility, and other constraints is crucial. Monitoring data exchanges and detecting potential violations are key challenges.
Emerging approaches, including federated governance models, AI-driven policy enforcement, and LLM-based automation, offer innovative ways to address these challenges.
In this context, we propose thesis concerning:
- Design a system for monitoring cross-organizational data sharing in a Data Mesh environment, ensuring compliance with agreements on privacy, performance, and visibility.
- Use LLMs to automatically generate Rego policies for the Open Policy Agent (OPA) from textual descriptions of agreements and constraints.
Integrate the generated policies into OPA to monitor REST-based data services in real time.
- Evaluate the system for accuracy, completeness, and effectiveness in detecting compliance violations.
Expected Outcomes:
- Prototype systems or frameworks supporting distributed Data Mesh governance.
- Guidelines and best practices for balancing domain autonomy and compliance.
- Evaluation metrics for monitoring quality, privacy, performance, and policy enforcement.
- Original contributions to research on scalable, automated, and AI-driven governance in decentralized data architectures.
Hide Description
Automated Policy Enforcement in Data Mesh: AI-driven ApproachesContact person
PIERLUIGI PLEBANIEmail:
pierluigi.plebani@polimi.itStudy course: Computer Science and Engineering
Web page:
https://plebani.faculty.polimi.it/Description
Description:
Data Mesh is an emerging paradigm for cross-organizational data sharing, emphasizing domain-oriented ownership, self-service data infrastructure, and decentralized governance. In multi-organization contexts, ensuring compliance with agreements on privacy, performance, visibility, and other constraints is crucial. Monitoring data exchanges and detecting potential violations are key challenges.
Emerging approaches, including federated governance models, AI-driven policy enforcement, and LLM-based automation, offer innovative ways to address these challenges.
In this context, we propose thesis concerning:
- Explore how AI techniques, including Large Language Models (LLMs), can support automatic policy enforcement in Data Mesh architectures.
- Automate tasks such as data classification, tagging, and compliance verification (privacy, access control, retention).
- Investigate mechanisms for explainability and accountability to ensure transparency of automated governance.
Expected Outcomes:
- Prototype systems or frameworks supporting distributed Data Mesh governance.
- Guidelines and best practices for balancing domain autonomy and compliance.
- Evaluation metrics for monitoring quality, privacy, performance, and policy enforcement.
- Original contributions to research on scalable, automated, and AI-driven governance in decentralized data architectures.
Hide Description
Automatic Compliance in Choreographic BPM Protocols through BlockchainContact person
PIERLUIGI PLEBANIEmail:
pierluigi.plebani@polimi.itStudy course: Computer Science and Engineering
Web page:
https://plebani.faculty.polimi.itDescription
Description:
Business Process Management (BPM) is increasingly moving toward inter-organizational scenarios, where multiple organizations collaborate through choreographic protocols. In these contexts, ensuring compliance the adherence to regulations, contractual obligations, and governance policies becomes crucial.
Emerging technologies offer innovative tools to address this challenge. In particular, Blockchain, enabling transparency, immutability, and automated compliance checks through smart contracts.
In this context, we propose thesis concerning:
- Translation of compliance rules into smart contracts for distributed auditing.
- Analysis of scalability, transparency, and privacy trade-offs.
- Application to a case study (supply chain, healthcare, logistics)
Expected Outcomes:
- A prototype software tool supporting compliance in choreographic protocols.
- A critical evaluation of the chosen technology for blockchain.
- An original contribution to the study of distributed governance in business processes.
Hide Description
LLMs as "Compliance Advisors" for Choreographic BPM ProtocolsContact person
PIERLUIGI PLEBANIEmail:
pierluigi.plebani@polimi.itStudy course: Computer Science and Engineering
Web page:
https://plebani.faculty.polimi.itDescription
Description:
Business Process Management (BPM) is increasingly moving toward inter-organizational scenarios, where multiple organizations collaborate through choreographic protocols. In these contexts, ensuring compliance the adherence to regulations, contractual obligations, and governance policies becomes crucial.
Emerging technologies offer innovative tools to address this challenge. In particular, Large Language Models (LLMs), which can support the extraction and verification of compliance rules from complex regulatory texts.
In this context, we propose thesis concerning:
- Use of language models to extract constraints from regulatory texts.
- Automatic analysis of process execution logs to detect violations.
- Evaluation of reliability and explainability of LLMs in governance scenarios.
- Case study on banking protocols, GDPR, or healthcare compliance.
Expected Outcomes:
- A prototype software tool supporting compliance in choreographic protocols.
- A critical evaluation of the chosen technology for LLMs.
- An original contribution to the study of distributed governance in business processes.
Hide Description
Syntesis of multi-FPGA acceleratorsContact person
SERENA CURZELEmail:
serena.curzel@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Description
Description:
High-Level Synthesis tools generate high-quality FPGA accelerators starting from software code. However, there is little automation available to users of large-scale multi-FPGA systems to help them deploy an application exploiting multiple FPGAs. This thesis will explore methods to partition and map FPGA designs, taking care of both communication and computation efficiency.
Hide Description
Advanced Topics in PhotonicsContact person
FRANCESCO MORICHETTIEmail:
francesco.morichetti@polimi.itStudy course: Automation Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Andrea MelloniWeb page:
https://photonics.deib.polimi.it/Description
Description:
The Photonic Devices group of Politecnico di Milano, leaded by Proff. Andrea Melloni and Francesco Morichetti has many openings for master thesis.
The group is involved in fundamental as well as applied research in photonics, from waveguide technology to circuit design, from industry applications to new concepts in the field of communication (fiber and free space), sensing, circuit control, testing and photonic computing.
Activities range from numerical analysis to experimental characterizations to fabrication in clean room. The group works in a multidisciplinary scenario merging photonics with electronics, control systems, computer science, machine learning, ….
We have different open positions for master thesis on various topics:
- Programmable Photonic Circuits
- Photonic computing
- Wireless optics and microwave photonics
- Photonic sensors and LIDAR
- Photonic MEMS
- Photonics for quantum systems
- Silicon (oxy)carbide technology, waveguides and passive/active devices conception and fabrication
- Testing techniques and tools for volume and on-wafer level
- Electronics and algorithms for circuit control
- Micro and nano fabrication in clean room (Polifab)
Students interested in a thesis activity with us should have some background in one of the following areas: photonics, telecommunication, electronics, RF systems, electromagnetism and physics, material science, engineering, control systems, computer science.
About us
https://photonics.deib.polimi.it
https://www.polifab.polimi.it
If you are interested or require more information contact andrea.melloni@polimi.it and francesco.morichetti@polimi.it.
Hide Description
Guided Time Series Generation for Alternative Scenarios ExplorationContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Web page:
https://www.deib.polimi.it/eng/people/details/1696874Description
Description:
Synthetic time series (TS) have recently improved in areas like data augmentation, forecasting, and filling missing data. They are useful for creating new or "what if" scenarios to test ideas and algorithms safely before using them in real situations, especially when making decisions under uncertainty. In finance, small changes in economic factors, interest rates, or investor mood can significantly affect asset prices and investments due to market volatility. Synthetic TS help analysts and risk managers explore different economic situations to evaluate their effects on investment strategies and risk. This thesis focuses on combining advanced neural networks with effective data transformations, such as the Fast Continuous Wavelet Transformation (FCWT) [1]. This combination aims to improve how we understand and use information from time-series data, enabling better creation of alternative scenarios to support decision-making.
Hide Description
Learning Dynamics from Small Data with Neural ODEs and Physical LawsContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Web page:
https://www.deib.polimi.it/eng/people/details/1696874Description
Description:
Standard deep learning fails on the small datasets common in science. This thesis proposes a new paradigm: meta-learning for few-shot system identification. We train a Neural ODE not to solve one task, but to learn how to learn from many related physical systems. By learning this shared structure, the model acquires a physical prior that allows it to accurately identify the dynamics of a novel system from only a handful of data points. This framework overcomes the small data bottleneck, enabling rapid and robust modeling where data is prohibitively expensive or difficult to acquire.
Hide Description
Adaptive Multimodal Reasoning via Dynamic Merging of Pre-trained ModelsContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Web page:
https://www.deib.polimi.it/eng/people/details/1696874Description
Description:
Contemporary AI systems often rely on static, monolithic architectures, which struggle to adapt to the dynamic and heterogeneous nature of real-world problems. The inability to dynamically adjust their reasoning process based on input complexity or modality leads to suboptimal performance and inefficiency. This thesis aims to introduce a novel framework for adaptive reasoning that intelligently orchestrates a committee of specialized, pre-trained models to solve complex multimodal tasks.
Hide Description
Tabular Foundation Models for Concept DriftContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Web page:
https://www.deib.polimi.it/eng/people/details/1696874Description
Description:
The advent of Foundation Models has initiated a paradigm shift in machine learning, with recent efforts extending their success to tabular data. These Tabular Foundation Models (TFMs) promise powerful, generalizable representations by pre-training on vast and diverse datasets. However, their static, pre-trained nature renders them inherently vulnerable to concept drift, a phenomenon where the underlying statistical properties of data change over time. This limitation is particularly pronounced in time-series applications, leading to significant performance degradation. This thesis addresses this critical gap by proposing a novel framework for Interactive Adaptive Learning. Our approach aims to pave the way for the robust and reliable deployment of TFMs in dynamic, real-world environments.
Hide Description
Analyzing Political Echo Chambers and Misinformation Dynamics on Truth SocialContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Automation Engineering, Computer Science and Engineering, Telecommunications Engineering
Web page:
https://pierri.faculty.polimi.itDescription
Description:
This study investigates the formation and behavior of political echo chambers and the spread of misinformation on Truth Social, a platform known for its alignment with right-wing discourse. Using computational social science methods, including network analysis and natural language processing (NLP), the research examines how information circulates within ideologically homogeneous communities and how narratives evolve in response to political events. The goal is to deepen our understanding of alternative social media ecosystems and their impact on political polarization and information integrity in the digital public sphere.
Hide Description
Digital Twin Framework for Intrabody RF Wireless CommunicationsContact person
SILVIA MURA, FRANCESCO LINSALATAEmail:
silvia.mura@polimi.it, francesco.linsalata@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electrical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Maurizio MagariniDescription
Description:
This thesis presents a digital twin framework for modeling and optimizing intrabody RF wireless communications across frequencies including millimeter-wave (mmWave) and terahertz (THz) bands. By integrating detailed anatomical and physiological data, the framework accurately replicates the complex electromagnetic environment within the human body. This enables predictive simulations of signal propagation, addressing key challenges such as attenuation, scattering, and absorption. The digital twin provides a non-invasive platform for designing and optimizing intrabody wireless systems, facilitating the advancement of high-performance networks for biomedical applications like implantable devices and real-time health monitoring.
Hide Description
Contributions to the development of a 5G snifferContact person
EUGENIO MOROEmail:
eugenio.moro@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Marco MezzavillaDescription
Description:
We are looking for a master student interested in pursuing their thesis on the topic of 5G networks. More specifically, the thesis is about contributing to the development of a 5G sniffer. In broad terms, a 5G sniffer is a device capable of decoding the transmissions between gNB and UEs.
The ideal candidate has the following skills:
- Proficient with C/C++
- Familiar with CMake
- Knowledgeable about wireless communication systems, specifically mobile radio networks
Bonus points if you have:
- experience with SDR
- experience with OpenAirInterface
This thesis will be carried out in collaboration with a team from KAIST university, but it does not require you to move to Korea.
Hide Description
Examining the Dynamics of Online Hate Speech in the Context of the Israel-Palestine ConflictContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Web page:
pierri.faculty.polimi.itDescription
Description:
This study investigates the dynamics of online hate speech related to the Israel-Palestine conflict by analyzing social media discourse across major platforms. Using NLP and discourse analysis, it explores how hate speech fluctuates in response to key events and assesses the effectiveness of platform moderation. The goal is to better understand digital polarization in geopolitical conflicts.
Hide Description
Solving image matching in Computer Vision exploiting permutation matricesContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Luca MagriDescription
Description:
A prominent problem in Computer Vision is multi-image matching, where the task is to find matches (i.e, coordinates of image points that derive from the same 3D point) across multiple images. This can be formalized as an optimization problem (called “synchronization”) over permutation matrices, and solved via eigenvalue decomposition. The goal of this thesis is to develop a new method for multi-image matching based on permutation matrices.
Reference: F. Arrigoni, A. Fusiello. Synchronization problems in Computer Vision with Closed-form solutions. IJCV 2020 https://link.springer.com/article/10.1007/s11263-019-01240-x
Hide Description
Identifying degenerate configurations in 3D reconstruction problems in Computer VisionContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Luca MagriDescription
Description:
3D reconstruction is an important problem in Computer Vision grounded in projective geometry. It is known that 3D reconstruction from two images can not be performed in some degenerate cases, for example when the 3D scene is a critical surface (a hyperboloid, a cone, two planes) and the cameras lie on that surface. The goal of this thesis is to develop a new method for detecting degenerate configurations in practice. A good mathematical background is preferred but not mandatory.
References: https://openaccess.thecvf.com/content/CVPR2022/papers/Fan_On_the_Instability_of_Relative_Pose_Estimation_and_RANSACs_Role_CVPR_2022_paper.pdf
https://link.springer.com/article/10.1007/s10851-019-00908-w
Hide Description
Wi-Fi Systems for Sport Racing CommunicationContact person
LUCA REGGIANIEmail:
luca.reggiani@polimi.itStudy course: Telecommunications Engineering
Description
Description:
Wi-Fi is the well-known family of wireless local area network protocols based on the IEEE 802.11 standard. The recent releases 6 and 7 have reached peak data rates around 10 and 46 Gbps respectively in bands at 2.4, 5, 6 GHz and they are widely deployed for short-medium range communication in indoor and outdoor environments. In cooperation with Marelli Motorsport, this thesis explores the use of Wi-Fi 6 for sport racing communication applications, in outdoor scenarios with medium-range data transfer and affected by interference of concurrent wireless systems. The study involves simulative and experimental activities.
Hide Description
Signal Processing and Learning for Near-Field 6G SystemsContact person
MAROUAN MIZMIZIEmail:
marouan.mizmizi@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Description
Description:
As 6G networks evolve to use higher frequencies and larger MIMO arrays, communication will increasingly occur in the near-field, unlike today’s far-field systems. This shift has significant implications for how channels are modeled, estimated, and how communication systems are designed. Near-field communication introduces unique challenges, such as the need for new channel models that account for spatial variations over short distances and more complex, yet critical, channel estimation techniques. Precoding and equalization strategies must also be reimagined to address the non-uniform distribution of signals and the potential for increased interference in dense network environments.
This thesis will explore these challenges by developing advanced signal processing techniques tailored for near-field conditions in 6G networks. It will focus on improving channel modeling, estimation, precoding, and equalization for high-frequency, dense scenarios. Additionally, the research will investigate the use of machine learning to enhance these techniques, offering more adaptive and robust solutions. The goal is to make a significant contribution to 6G technology by addressing the critical challenges of near-field communication with both traditional and machine learning-based approaches.
Hide Description
Shaping the Future of 6G Networks: Digital Twins, AI, and Advanced Wireless EmulationContact person
FRANCESCO LINSALATAEmail:
francesco.linsalata@polimi.itStudy course: Automation Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Eugenio MoroDescription
Description:
The increasing complexity of 6G networks requires advanced tools for accurate modeling, real-time optimization, and intelligent management. Digital Network Twins (DNTs) represent a transformative approach, providing precise digital replicas of physical networks to enable predictive analysis, performance optimization, and risk-free experimentation.
This thesis will explore the integration of AI and Machine Learning (ML) techniques with ray-based channel modeling to enhance the accuracy and adaptability of DNTs. By leveraging data-driven methodologies, the research aims to develop a framework capable of dynamically simulating and optimizing next-generation wireless networks.
This project is suitable for students with a strong interest in AI-driven network management, wireless communications, and the evolution of 6G technologies. While prior experience in simulation, emulation, or optimization is beneficial, it is not a prerequisite, making this an excellent opportunity to develop expertise in these areas.
Hide Description
Exploring Security Vulnerabilities in the OpenTitan Framework: A Comprehensive Analysis and Testing ApproachContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia LazzeriWeb 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
Analysis of NASA's Core Flight System (cFS) Framework: An Application Case Study for Electro-Optical Payload Management on CubeSatsContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
in collaboration with TDS-Space and the European Space AgencyWeb page:
https://cassano.faculty.polimi.it/Description
Description:
The Core Flight System is an advanced software architecture based on the mission operation services of the CCSDS (Consultative Committee for Space Data Systems). Developed by NASA, this software is designed to support space operations and system integration for various aerospace missions, thanks to its modularity that allows updates similar to smartphone applications. NASA's cFS is a software framework intended to support space operations and system integration across diverse aerospace missions. Thesis/Internship Objectives: The thesis and internship activities are designed to investigate the effectiveness of NASA's Core Flight System (cFS) in managing electro-optical payloads on CubeSat platforms. This evaluation specifically focuses on the performance of the cFS when operational on a representative target computing platform for data handling and processing, developed by TSD-Space. Specifically, the focus will be on implementing a cFS application that manages image acquisition operations, integrating monitoring and control services performed through the cFS framework and the C programming language.
Hide Description
Real-Time Onboard 2D Image Correlation System for Estimating and Compensating Instantaneous Focal Plane Motion: An FPGA implementationContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
in collaboration with TDS-Space and the European Space AgencyWeb page:
https://cassano.faculty.polimi.it/Description
Description:
The activities will be conducted in the framework of the FOPAC project, led by TSD-Space and founded by the Italian Space Agency. The FOPAC’s goal is to develop an active motion control system for the focal plane of electro-optical payloads adopted in Earth observation applications to enhance image quality in terms of spatial resolution and Signal-to-Noise Ratio (SNR). These platforms often feature telescopes with limited focal length and aperture, and less accurate attitude control systems. FOPAC can be utilized for implementing active focal plane stabilization techniques and pixel-shifting acquisition mode. Active stabilization mitigates motion blur effects due to platform instability and enables the adoption of advanced Time Delay Integration (TDI) techniques for SNR improvements. The FOPAC project incorporates the implementation of real-time onboard 2D image correlation algorithms to estimate the instantaneous motion of the focal plane and stabilize it. This compensates for the limited attitude control performance of small platforms, which significantly impacts image quality. Focal plane stabilization, capable of compensating for platform jitter—a source of instability—allows for increased exposure times, thereby enhancing the SNR. Alternatively, with the same exposure time, it reduces motion blur, thus improving the Modulation Transfer Function (MTF). The primary activity of the candidate will be to explore and survey different approaches in 2D image correlation and motion estimation, focusing on new algorithms based on AI. The selected algorithm will be accelerated on an FPGA platform for real-time execution.
Hide Description
Real-time onboard satellite image registration for Digital TDI: An FPGA ImplementationContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
In collaboration with TDS-Space and the European Space AgencyWeb page:
https://cassano.faculty.polimi.it/Description
Description:
In the field of Earth observation, the digital Time-Delay Integration (TDI) acquisition technique is crucial for enhancing the quality of satellite images. This method utilizes electro-optical payloads to capture sub-stripes or partially overlapping images. By digitally summing the signals encoded in the pixels of different sub-stripes that correspond to the same ground point, it
is possible to reduce uncorrelated noise and significantly increase signal amplitude. This process leads to a substantial improvement in the signal-to-noise ratio (SNR) and, consequently, the overall quality of the image. However, the digital TDI technique requires capturing multiple images of the same sample, thereby increasing the resources needed for data storage and transmission to ground stations. By performing digital TDI operations directly onboard the spacecraft, it is possible to drastically minimize the volume of data to be stored and maximize the efficiency of bandwidth use for downlink.
Digital TDI activity essentially includes image alignment, which is a prerequisite to ensure that the sum of signals occurs exclusively among pixels corresponding to the same ground sample. During the internship at TSD-Space, the candidate will explore various methods for digital image registration, including advanced approaches that incorporate artificial intelligence. The selected algorithm will be accelerated on an FPGA platform for real-time execution, significantly enhancing the effectiveness of the TDI process.
Hide Description
5G-V2X positioning of a vehicle in a tunnelContact person
MATTIA BRAMBILLAEmail:
mattia.brambilla@polimi.itStudy course: Telecommunications Engineering
Other members of the research group:
Monica NicoliDescription
Description:
This thesis aims to investigate signal processing and learning algorithm to localize a vehicle traveling inside a tunnel without GPS coverage, using an on-site infrastructure of 5G V2X MIMO antennas.
First, the student will consider a ray tracing software tool for modeling the propagation of a wireless signal in the tunnel. Then, the student will develop algorithms exploiting the features of the received signal for positioning. At the end, if possible, the student should evaluate the feasibility of applying the "simulated-data-driven algorithm" on real measurements, assessing the portability from simulation to the real world of the algorithm.
The thesis is framed within a project with Movyon - Autostrade per l'Italia.
Hide Description
Randomized based digital systems verificationContact person
FABRIZIO FERRANDIEmail:
fabrizio.ferrandi@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
In collaboration with Analog Device, Milano ItalyDescription
Description:
Digital design verification involves testing and confirming that a digital design or system is correct and functions properly before it is launched or used. The main objective of this process is to find and fix any design errors or bugs, ensuring that the system operates as intended in various conditions and scenarios.
Randomization is commonly used to generate test stimuli for digital verification-based approaches. It begins with constraints specifying the allowed values and utilizes a seed to generate the stimuli. In current industry practice, there is no standard approach to randomization, which often leads to the same constraints producing different stimuli across different simulation engines.
The thesis focuses on defining a portable randomizer that can connect to existing simulation engines, both free and commercial. This randomizer will intercept calls for randomization and deliver unique stimuli based solely on specified constraints and a seed value. The activity will be conducted in collaboration with the digital verification group at Analog Devices in Milan, Italy.
Hide Description
Rete Ferroviaria ItalianaContact person
NICOLA LUSARDIEmail:
nicola.lusardi@polimi.itStudy course: Electronics Engineering
Other members of the research group:
Fabio Garzetti, Angelo GeraciDescription
Description:
DigiLAB collaborates with other laboratories from Politecnico di Milano and various Italian universities on research and development projects with Rete Ferroviaria Italiana (RFI) to create secure embedded systems for railway applications.
DigiLAB focuses on developing electronic boards based on FPGAs, SoCs, and next-generation processors, necessary for executing the secure code of RFI’s railway algorithms. The large volume of exchanged data requires high-speed communication interfaces (such as Ethernet and HDMI). Additionally, using SoCs is crucial, allowing the creation of embedded systems with dedicated hardware integrated directly into FPGAs.
Hide Description
European XFELContact person
NICOLA LUSARDIEmail:
nicola.lusardi@polimi.itStudy course: Electronics Engineering
Other members of the research group:
Fabio Garzetti, Angelo GeraciDescription
Description:
DigiLAB, together with other laboratories from Politecnico di Milano, collaborates with European XFEL, DESY, the University of Heidelberg, the University of Bergamo, and INFN to develop the DSSC X-Ray Imager (DSSC).
https://fe.desy.de/fec/projects/photon_science/dssc_x_ray_imager/
https://www.xfel.eu/
DSSC is a specialized 1-Mpixel X-ray camera for images generated by EuXFEL. It uses a nonlinear sensor (DepFET) capable of capturing up to 8,000 frames per second. DepFET sensors are managed in groups of 4 kpixels by dedicated ASICs that digitize the acquired data.
DigiLAB’s challenge is developing the firmware for the data acquisition (DAQ) system, which must handle data flows in the tens of Gbps. Specifically, the 1-Mpixel camera is divided into four quadrants, each featuring four Spartan-6 FPGAs for managing 16×4 ASICs, and a Kintex-7 FPGA for transmitting data to EuXFEL servers.
https://ieeexplore.ieee.org/document/10210050
Currently, measurements are being taken on the second version of the DSSC.
https://ieeexplore.ieee.org/document/10657049
Hide Description
Elettra Sincrotrone di TriesteContact person
NICOLA LUSARDIEmail:
nicola.lusardi@polimi.itStudy course: Electronics Engineering
Other members of the research group:
Fabio Garzetti, Angelo GeraciDescription
Description:
DigiLAB has a decade-long collaboration with the “Instrumentation & Detectors Labs” at Elettra Sincrotrone di Trieste S.C.p.A.
https://www.elettra.eu/it/index.html
As part of this collaboration, we develop the firmware required for the readout electronics of Cross Delay-Line (CDL) detectors and instrumentation for pump-and-probe time-resolved experiments.
CDL detectors are systems used in synchrotrons for 3D imaging, capable of providing each frame with spatial details at a resolution of tens of micrometers in x-y coordinates, while simultaneously offering time-of-arrival (t) information for each pixel with a precision of a few tens of picoseconds. The main challenge of these systems, besides high resolution, is minimizing dead-time (in the nanosecond range), ensuring the ability to acquire signals spaced only a few nanoseconds apart without degrading image quality.
https://ieeexplore.ieee.org/document/9855502
Pump-and-probe experiments are essential for material analysis. In synchrotrons, a laser is used as the “pump” source to excite the sample, while X-rays from the synchrotron serve as the “probe,” scanning different energy levels. The challenge is detecting the pulses emitted by the tested sample and correctly correlating them with the synchrotron and laser triggers, enabling beamline scientists to perform data analysis.
https://www.mdpi.com/1424-8220/24/16/5233
Hide Description
Intrabody Communication: Exploring different technologies for Biomedical Data TransmissionContact person
SILVIA MURAEmail:
silvia.mura@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Maurizio MagariniDescription
Description:
Intrabody communication (IBC) is a promising technique for secure and energy-efficient data transmission within the human body, with applications in biomedical devices, wearable sensors, and implantable systems. The exploration of different transmission methods, including terahertz (THz) waves, galvanic coupling, and ultrasounds, presents an opportunity to optimize performance for diverse medical and health monitoring applications. This research seeks to analyze the feasibility and efficiency of these communication techniques through a combination of simulations and experimental validation.
Hide Description
UAV Trajectory Planning and Synchronization for Efficient Search and Rescue OperationsContact person
SILVIA MURAEmail:
silvia.mura@polimi.itStudy course: Automation Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Maurizio Magarini, Francesco Linsalata, Davide ScazzoliDescription
Description:
This research focuses on advanced trajectory planning and synchronization techniques for Unmanned Aerial Vehicles (UAVs), commonly known as drones, in search and rescue operations. By optimizing flight paths, real-time communication, and coordinated movement, the study aims to enhance efficiency in locating and assisting individuals in distress. The thesis aims to integrate path optimization algorithms, obstacle avoidance mechanisms, and cooperative control strategies to ensure seamless multi-drone collaboration, even in complex and unpredictable environments. The findings will contribute to improving autonomous drone operations for emergency response and disaster management.
Hide Description
Development of a Multi-Core RISC-V SoC for Harsh EnvironmentsContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Dr. Luigi Dilillo (University of Montpellier, France)Web page:
https://cassano.faculty.polimi.it/Description
Description:
Radiation-induced effects, such as Single Event Upsets (SEU) and Total Ionizing Dose (TID), pose significant challenges for electronic systems operating in space and other harsh environments. This internship will focus on extending an existing RISC-V-based System-on-Chip to a multi-core architecture in order to enhance performance and fault tolerance against radiation effects. The work involves modifying the existing VHDL design to support multiple cores, implementing inter-core communication and synchronization, and integrating fault-mitigation techniques. Candidates should have experience in VHDL, processor architecture, C programming, and microcontrollers.
Hide Description
Exploiting GPU Microarchitectures for Novel Hardware Attacks: Investigating Memory Leaks and Arbitrary Code Execution Without Physical AccessContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia LazzeriWeb 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
Development of a Linux Kernel Module for Fault Injection in ProcessorsContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Elia LazzeriWeb page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis aims to develop a Linux kernel module enabling fault injection into selected processes or programs running on a processor. The tool will simulate adverse conditions, such as ionizing radiation, that affect memory and computation. The solution may involve a kernel module, a small fork, or a patch to the Linux kernel, allowing controlled or random fault injection for automated and repeatable testing. The system will help evaluate application resilience to faults and will include a configurable logging mechanism at multiple levels to track performance and events.
Hide Description
Optimization Algorithms for Continual Deep LearningContact person
FABRIZIO PITTORINOEmail:
fabrizio.pittorino@polimi.itStudy course: Computer Science and Engineering
Web page:
https://www.deib.polimi.it/eng/people/details/1696874Description
Description:
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.
Hide Description
Auditing social media platforms for Teen usage via sockpuppet accountsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Web page:
pierri.faculty.polimi.itDescription
Description:
This thesis explores the use of automated and human-operated sockpuppet accounts to audit social media platforms’ policies and practices regarding teen users. It proposes a methodology for systematically testing age verification, exposure to harmful content, and engagement-driven algorithmic biases. By combining experimental digital ethnography with LLM-assisted content analysis, the research evaluates platform compliance, risk factors, and potential regulatory gaps.
Hide Description
Understanding memes with Large Language ModelsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Computer Science and Engineering
Web page:
pierri.faculty.polimi.itDescription
Description:
This thesis investigates the application of large language models (LLMs) to analyze internet memes, leveraging KnowYourMeme.com as a structured dataset. It proposes a framework for categorizing memes based on linguistic, visual, and cultural features, assessing their evolution, virality, and ideological impact. By applying natural language processing, semiotic analysis, and network modeling, the study aims to decode how memes propagate, adapt, and influence digital discourse.
Hide Description
Impact of dataset reuse in web and data miningContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Web page:
pierri.faculty.polimi.itDescription
Description:
Publicly shared datasets are crucial for research in web and data mining, yet their reuse remains understudied. Many conferences now feature dataset or resource tracks, highlighting the growing importance of data sharing. This project seeks to quantify dataset reuse by analyzing research papers, tracking dataset mentions, and identifying key trends in adoption. We aim to determine which datasets are most influential and what factors impact their reuse, such as accessibility and documentation. Understanding these patterns will inform best practices for dataset sharing and encourage greater transparency in research.
Hide Description
Exploiting Software Obfuscation to prevent Transient Execution Attacks in RISC-V processorsContact person
LUCA CASSANOEmail:
luca.cassano@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering
Other members of the research group:
Elia LazzeriWeb page:
https://cassano.faculty.polimi.it/Description
Description:
This thesis focuses on the design and development of software obfuscation engine to enhance the security of computing systems. The idea is to modify the control and data flows of a program in order to make Transient Execution Attacks like Spectre and Meltdown more difficult to be executed or even impossible.
Hide Description
Real-Time Heart Signal Anomaly Detection and Prevention Using Radar TechnologyContact 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:
Francesco LinsalataDescription
Description:
This thesis focuses on developing algorithms to detect anomalies in heartbeat signals captured by radar, with an emphasis on decoupling cardiac signals from respiration to enhance detection accuracy. The research incorporates Integrated Sensing and Communication (ISAC), enabling radar systems to seamlessly combine physiological monitoring with communication functionality.
The student will design and implement methods for detecting or preventing heart rate anomalies, such as arrhythmias, and validate their performance through measurement campaigns using real radar data. The ultimate goal is to deliver a low-complexity, real-time solution for healthcare applications and preventative monitoring.
Hide Description
Semantic Signal Reconstruction for Restoring Nerve Signal PropagationContact 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:
Maurizio MagariniDescription
Description:
This thesis explores a novel approach to restoring nerve function through semantic signal reconstruction, designed to act as a relay system for damaged nerves. By focusing on the "semantic meaning" of neural signals—the critical information necessary for physiological responses—this method bypasses the need to replicate the entire signal, enabling efficient and targeted signal propagation across injured sections. The work emphasizes low-complexity algorithms and real-time processing, making the approach suitable for implantable devices like cuff electrodes. These devices capture, interpret, and regenerate nerve signals, ensuring seamless communication despite nerve damage.
Hide Description
Simulating social media platforms with LLMsContact person
FRANCESCO PIERRIEmail:
francesco.pierri@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Web page:
pierri.faculty.polimi.itDescription
Description:
This thesis explores using LLMs to simulate social media interactions, enabling controlled studies of digital discourse, misinformation, and user behavior. It proposes a framework for generating artificial social media environments, analyzing emergent behaviors, and assessing realism. The research applies reinforcement learning and network analysis to model interactions, offering insights for policy-making, marketing, and cybersecurity. The study aims to enhance understanding of algorithmic influence, misinformation spread, and engagement patterns in online communities.
Hide Description
5G-enabled passive radar for detecting unauthorized dronesContact person
MAURIZIO MAGARINIEmail:
maurizio.magarini@polimi.itStudy course: Automation Engineering, Electrical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Davide ScazzoliDescription
Description:
The goal of the proposed master's thesis is to investigate the detection of unauthorized drones using a passive bistatic radar (PBR) system that leverages a 5G base station as the transmitter. This method utilizes existing infrastructure, making it a cost-effective solution for drone surveillance. The detection process involves modeling the communication environment between the 5G base station and the drone, and analyzing the signals reflected from the drone. A crucial aspect of the study is calculating the Radar Cross Section (RCS) to measure how the drone scatters radar signals. By combining signal analysis with RCS calculations, the system can detect small, agile drones in complex environments. The thesis evaluates the system's performance with a specific drone model under various conditions, demonstrating that integrating PBR with 5G networks can effectively detect and track unauthorized drones, enhancing security in critical areas.
Hide Description
Cellular infrastructure for weather monitoringContact person
DARIO TAGLIAFERRIEmail:
dario.tagliaferri@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Description
Description:
The thesis aims at studying the potential of the cellular infrastructure for the retrieval and monitoring of rain.
The goal of the work is to investigate the potential of the cellular network to be use for weather forecasting, towards greener environments.
The student will learn how to model and process the cellular signals to estimate the rain precipitation rate, and validate the simulations on experimental data acquired with a 5G cellular base station.
Hide Description
LEO satellites for communication and environment monitoringContact person
DARIO TAGLIAFERRIEmail:
dario.tagliaferri@polimi.itStudy course: Automation Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Francesco LinsalataDescription
Description:
Low Earth orbit (LEO) satellites, e.g., Starlink, will be a keystone in future 6G non-terrestrial wireless networks. In addition to the communication functionality, LEO fleets can also serve for accurate Earth monitoring, with applications ranging from infrastucture monitoring, public security, surveillance and environmental reconstruction.
The thesis consists of exploring the use of LEO satellites to jointly ensure broadcast connectivity while obtain an accurate environment mapping.
The work will involve practical activities within the framework of the PNRR project "RESTART".
Hide Description
AI for Cooperative Perception in Vehicular Radar NetworksContact person
DARIO TAGLIAFERRIEmail:
dario.tagliaferri@polimi.itStudy course: Automation Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Description
Description:
The road towards autonomous driving expects vehicles will be equipped with more and more heterogeneous sensors (cameras, lidars, radars, sonars) to perceive the surroundings. Cameras and lidars are the golden standard for autonomous driving, providing high-resolution representations of the environment, but they are sensitive to illumination levels and weather conditions (fog, rain,etc).
Radars, instead, are robust to adverse weather conditions, but are typically characterized by low spatial resolution in the description of the environment and they are currently used for limited purpose (emergency braking).
To improve the reliability of the perception of the environment, future autonomous vehicles are expected to mutually cooperate, exchanging local measurements and features to augment perception.
The thesis aims at using artificial intelligence for cooperative perception based on radars. In particular, the goal of the thesis is to investigate novel AI architectures for aligning, synchronizing and fusing radar images generated at the single vehicles. The proposed thesis can be specialized to generic distributed radar networks, e.g., made of drones.
Hide Description
Integrated sensing and communication for future 6G wireless systems in the upper mid-bandContact person
DARIO TAGLIAFERRIEmail:
dario.tagliaferri@polimi.itStudy course: Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Marco MezzavillaWeb page:
Pi-Radio website: https://www.pi-rad.io/homeDescription
Description:
Integrated Sensing and Communication (ISAC) systems combine wireless communication and radar sensing capabilities within the same infrastructure. These systems leverage shared hardware, spectrum, and signal processing techniques to perform tasks like data transmission and environmental sensing simultaneously. ISAC enables applications such as autonomous driving, smart cities, publoc safety and surveillance. It is acquainted that the next generation (6G) of wireless systems will use the so-called upper mid-band (6-24 GHz), currently not used for communication systems. The upper mid-band that offer large bandwidth over a wide range of frequencies, opening the striking possibility of centimeter-level environment sensing and unprecedented .
The thesis aims at addressing the innovative design of an ISAC systems in the future 6G frequencies. The student can choose to address the 6G ISAC from the methodological perspective , finding new ways to integrate communication and sensing over multiple disaggregated bandwidths, or developing custom signal processing algorithms that can be validated with in-lab experimental measurements with the Pi-Radio hardware.
Hide Description
DEFINIZIONE DI UN PROTOCOLLO PER PRODURRE UNA INTERFACCIA NEURALEContact person
MAURIZIO MAGARINIEmail:
maurizio.magarini@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electronics Engineering, Telecommunications Engineering
Other members of the research group:
Antonio CovielloDescription
Description:
La tesi si concentra sulla progettazione e la manifattura di una cuff electrode, un’interfaccia neurale da posizionare attorno a un nervo periferico per il campionamento del segnale nervoso. Lo studente sarà coinvolto in attività pratiche di laboratorio, con focus su biomateriali avanzati, tecniche di produzione, parziale progettazione circuitale e aspetti normativi. L’obiettivo è sviluppare un protocollo e un metodo di produzione che garantiscano un campionamento stabile ed efficace del segnale nel tempo.
Hide Description
REALIZZAZIONE HARDWARE DI UN CHIP IMPIANTABILE PER APPLICAZIONI MEDICALIContact person
MAURIZIO MAGARINIEmail:
maurizio.magarini@polimi.itStudy course: Biomedical Engineering, Electrical Engineering, Electronics Engineering, Telecommunications Engineering
Other members of the research group:
Antonio CovielloDescription
Description:
La tesi si concentra sulla progettazione e sviluppo hardware di un chip impiantabile, con focus sull’acquisizione e trasferimento dei segnali biomedicali. Lo studente sarà coinvolto nella scelta, configurazione e integrazione delle componenti elettroniche per la realizzazione del dispositivo. L’ottimizzazione si orienterà verso applicazioni mediche future sull’uomo, puntando a miniaturizzazione del segnale medicale, efficienza energetica e affidabilità a lungo termine.
Hide Description
Design of an eye tracking device exploiting mmWaves radarContact person
MAURIZIO MAGARINIEmail:
maurizio.magarini@polimi.itStudy course: Biomedical Engineering, Electrical Engineering, Electronics Engineering, Telecommunications Engineering
Other members of the research group:
Davide ScazzoliDescription
Description:
Eye-tracking capabilities have applications across a wide range of fields, including entertainment, medicine, and security. This thesis proposes the development of novel algorithms and devices that exploit mmWave radars for visual gaze analysis. The candidate will focus on both the practical aspects and the more abstract algorithmic components of the problem. The work includes an experimental phase involving the design of PCBs using commercial sensors to validate the developed algorithms. Additional extensions of this thesis will explore integrated sensing and communication using the same device.
Hide Description
Exploration of custom floating-point accelerators through evolutionary design methodsContact person
SERENA CURZELEmail:
serena.curzel@polimi.itStudy course: Computer Science and Engineering
Description
Description:
The open-source HLS tool Bambu, developed at Polimi, features a framework to customize floating-point encodings and obtain more efficient FPGA/ASIC accelerators (see https://hdl.handle.net/11311/1254959). The objective of this thesis is to use evolutionary design methods to select the best possible encoding for a given application under a given target accuracy.
Hide Description
Predictive Smart Irrigation Systems for Optimizing Plant Health Using Electrophysiological and Environmental Sensors with Deep LearningContact person
IMEN BEKKARIEmail:
imen.bekkari@polimi.itStudy course: Automation Engineering, Biomedical Engineering, Electrical Engineering, Electronics Engineering, Computer Science and Engineering, Telecommunications Engineering
Other members of the research group:
Imen BekkariDescription
Description:
This thesis aims to develop a predictive smart irrigation system by integrating electrophysiological sensors, electronic noses, and environmental sensors to monitor and analyze plant health. The model will optimize irrigation practices through neural networks to enhance water efficiency and promote sustainable agriculture in controlled and natural environments.
Hide Description
User-Centric IT Sustainability – Balancing Environmental Impact and User Experience in Adaptive Web ApplicationsContact person
MONICA VITALIEmail:
monica.vitali@polimi.itStudy course: Computer Science and Engineering
Description
Description:
The urgent need to reduce environmental impact has driven IT providers to reevaluate the management and allocation of physical resources for applications. While infrastructure-focused solutions have been prioritized, application developers also play a key role in achieving greener IT by designing applications that adapt dynamically to the availability of renewable energy. This flexibility, however, introduces a trade-off between environmental impact and user experience. For instance, a task could be executed in multiple ways, each with a different computational and energy cost, where the optimal version might be selected based on the availability of renewable energy sources. Such a selection, however, could influence the user experience and, subsequently, customer satisfaction.
This thesis seeks to understand this balance by taking a user-centric approach to IT sustainability, investigating which strategies can encourage consumers to actively support sustainable choices. Specifically, we aim to explore two key research questions:
- What factors influence users when they face a trade-off between sustainability and user experience?
- Can heightened sustainability awareness among consumers make them more receptive to minor reductions in user experience?
Our research will involve user studies to identify the actions and types of information that increase users’ willingness to accept slight QoS trade-offs for sustainability gains. By examining user preferences and the impact of sustainability-awareness strategies, this thesis will provide insights into how consumers can be empowered to support greener applications.
Hide Description
Balancing Privacy, Utility, and Sustainability: A Comparative Analysis of Synthetic Data Generation and Traditional Privacy-Preserving TechniquesContact person
MONICA VITALIEmail:
monica.vitali@polimi.itStudy course: Computer Science and Engineering
Description
Description:
Generative AI enables the creation of synthetic datasets that can address privacy concerns in sensitive data contexts, such as healthcare, finance, and social science research. By generating artificial data that mimics real data distributions without disclosing individual data points, these synthetic datasets promise a powerful alternative to traditional privacy-preserving methods like differential privacy and anonymization. However, this approach comes at an environmental cost. The creation and training of generative models require substantial computational resources, leading to significant energy consumption and, consequently, a notable carbon footprint. At the same time, traditional privacy-preserving methods are not without energy demands, particularly when applied to large and complex datasets. Thus, understanding the trade-offs in energy consumption, utility, and privacy risk between these approaches is crucial for sustainable and effective data privacy practices.
The primary objective of this thesis is to conduct a comprehensive comparison of the environmental and practical implications of synthetic dataset generation and traditional privacy-preserving techniques. This study will focus on three key areas: (1) energy consumption, measured through computational load and resource demands; (2) "fit to use," or the utility and fidelity of the datasets in maintaining representational accuracy; and (3) the risk profile, assessing each method’s effectiveness in safeguarding privacy.
The findings of this thesis will contribute to the emerging discourse on sustainable AI by providing data-driven insights into the environmental impact of privacy-preserving data practices. Given the increasing adoption of AI in sensitive fields and the rising concerns over its carbon footprint, this research aims to offer guidance on balancing data utility, privacy, and environmental responsibility.
Hide Description
AI’s Hidden Footprint: What It Really Takes to Power Large Language Models Like ChatGPTContact person
MONICA VITALIEmail:
monica.vitali@polimi.itStudy course: Computer Science and Engineering
Description
Description:
Large Language Models (LLMs) like ChatGPT are increasingly integrated into our daily lives, from assisting with information retrieval and content creation. As their popularity grows, so does the demand for training these models on massive datasets and frequently updating them in response to the evolving nature of knowledge. However, the computational requirements for training and updating LLMs results in considerable energy consumption. This raises concerns over the sustainability of data centers, especially as they strive to achieve carbon neutrality in the coming years.
Unlike traditional machine learning models, LLMs impose a continuous environmental impact that extends beyond the training phase. The inference phase, where models generate responses in real-time to meet high user demand, is similarly energy-intensive, often consuming as much or more power as training due to the computational complexity of each generated response. A balance between LLMs performance and sustainability needs to be studied to mitigate their environmental impact.
This thesis aims to systematically analyze the energy consumption patterns of popular LLMs, identifying key factors that drive their computational and environmental costs. We will evaluate the impact of various fine-tuning techniques on power consumption, investigating the trade-offs between model accuracy, efficiency, and resource demands.
Additionally, this study will develop a predictive model to estimate the energy consumption of both the training and inference phases for LLMs. This model aims to provide actionable insights to model developers and end-users alike, fostering awareness about the environmental footprint of using LLMs and encouraging more sustainable practices in AI deployment. Ultimately, this research will contribute to a better understanding of how we can optimize LLM usage while minimizing their ecological impact, helping to align technological advancement with environmental responsibility.
Hide Description
Empowering Sustainable Applications to Reduce Data Center Environmental ImpactContact person
MONICA VITALIEmail:
monica.vitali@polimi.itStudy course: Computer Science and Engineering
Description
Description:
As the demand for computational resources accelerates, the environmental impact of data centers is becoming increasingly significant. Efforts to address this impact have largely focused on infrastructural solutions, such as improving data center cooling efficiency, optimizing hardware, and transitioning to renewable energy sources. However, this infrastructure-centric approach overlooks an essential dimension: application design and development. Applications themselves, if designed with sustainability in mind, can play a pivotal role in reducing resource consumption and enhancing data center efficiency. This thesis proposes a framework that engages application designers and developers in creating adaptive, sustainable applications to contribute to greener, more environmentally-friendly cloud ecosystems.
This research introduces a novel architecture aimed at embedding sustainability features directly into cloud-native application components and provides a pathway for applications to dynamically adapt their workflows in response to environmental and resource constraints, enabling real-time adjustments based on available energy sources, data center load, and sustainability targets.
The proposed framework will focus on integrating sustainability as a core element in the design and development phases. Specifically, we will explore techniques for embedding adaptability features into application components, allowing them to modify their operations based on environmental conditions, such as shifting processing loads during times of low energy availability or optimizing computational intensity during peak periods. This approach moves beyond static, infrastructure-focused sustainability measures, paving the way for applications that can actively contribute to environmental goals.
Hide Description
Cascaded Foundation Models for the Segmentation of Abdominal OrgansContact person
ANDREA MOGLIAEmail:
andrea.moglia@polimi.itStudy course: Biomedical Engineering, Computer Science and Engineering
Other members of the research group:
Pietro Cerveri; Luca MainardiWeb page:
https://www.deib.polimi.it/eng/people/details/1991524Description
Description:
Multimodal large language models (MLLMs) were proposed over the last two years to generate text output from a prompt combining different input types, for instance, text and images. If applied to medicine, they are expected to generate new applications to support clinicians, e.g. in the decision-making when a medical doctor needs advice from a more experienced colleague. Some MLLMs specific to the medical domain were developed, like LLaVA-Med, PeFoMed, and Med-Flamingo. However, those MLLMs can only be used for visual question answering but not other tasks, like image segmentation, which is a critical component in clinical practice, facilitating accurate diagnosis, treatment planning, and disease progress monitoring. Segment Anything Model (SAM) was developed as the first foundational model for the segmentation of natural images. SAM automatically generates segmentation masks by selecting a point, drawing a bounding box around objects to segment, or even using a text prompt. SAM was not specifically developed for medical images, e.g., computed tomography scans, which are traditionally more challenging. Initial studies have shown that its performance was lower than that of deep learning models in the segmentation task of medical images. However, the recent release of MedSAM, a fine-tuned version of SAM on an unprecedented quantity and variety of datasets for medical images, outperforming state-of-the-art deep learning models, disrupted the field. Medical SAM 2, based on SAM 2, outperformed deep learning models, SAM-based ones, and even MedSAM in medical image segmentation.
This thesis work aims to combine MLLMs and foundation models like MedSAM/ Medical SAM 2 to build a tool enabling users to type a prompt like “Segment the pancreas” and visualize the segmentation mask. This thesis will build upon a recent research work at the Politecnico di Milano on a fine-tuned MLLM outputting a bounding box for the localization of the pancreas, given a question like “Where is the pancreas”. The automatically generated bounding box could serve as a prompt for MedSAM/Medical SAM 2 to segment the abdominal organs. Since current MLLMs need improvements in the localization of medical organs, the student(s) will need to optimize the MLLM architecture, for instance by replacing the vision encoder with a novel component, like MedNeXt .
Hide Description
Machine Learning for Quantum ComputingContact person
MAURIZIO FERRARI DACREMAEmail:
maurizio.ferrari@polimi.itStudy course: Computer Science and Engineering
Other members of the research group:
Riccardo Nembrini; Riccardo Pellini; Gloria Turati; Paolo CremonesiWeb page:
https://mauriziofd.github.io/Description
Description:
Quantum Computing (QC) has the potential to shape the future of high performance computing and Machine Learning (ML) applications thanks to the speed-ups it promises for certain tasks. However, ML techniques can also help to overcome some of QC's current technological limitations. This research aims to study how QC and ML can be best combined in practice and how they can be of help to each other. Examples of topics are the development of ML methods to design and optimize quantum circuits for specific architectures.
Hide Description
Good graphs versus bad graphs in 3D reconstructionContact person
FEDERICA ARRIGONIEmail:
federica.arrigoni@polimi.itStudy course: Computer Science and Engineering
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. A strong mathematical background is preferred but not mandatory. Reference: F. Arrigoni, A. Fusiello, T. Pajdla. A direct approach to viewing graph solvability. ECCV 2024
Hide Description
