Junior assistant Professor (fixed-term) at the University of Bologna since 2024. Her research activity lies at the interface between Theoretical and Mathematical Physics and the theory of Machine Learning and Deep Learning, using tools of Physics of Complex and Disordered Systems. Her research interests are mainly focused on the role played by data in Machine Learning problems, with a special focus on all those scenarios characterized by data scarcity. She is currently interested in the study and analysis of fundamental paradigms for training deep neural networks, such as transfer and self-supervised learning. She contributes to the application of Machine Learning and Deep Learning algorithms in the medical field as co-founder and director of the Research and Development department of SynDiag s.r.l., a spin-off of the Politecnico di Torino whose mission is the automatic detection of ovarian lesions from ultrasound videos.
Education
- 2012: Bachelor's Degree in Physical Engineering from Politecnico di Torino.
- 2014: Master's Degree in Physics of Complex Systems from Politecnico di Torino and Université Pierre et Marie Curie (Paris VI).
- 2018: PhD in Theoretical Physics applied to Machine Learning from Politecnico di Torino, with the dissertation titled "Statistical Physics of Neural Systems," supervised by Prof. Riccardo Zecchina.
Academic Career
- 2018: Research Fellow at Politecnico di Torino.
- 2019-2021: Post-Doctoral Researcher at Institut de Physique Théorique (IPhT) and École polytechnique fédérale de Lausanne (EPFL) in the groups led by Prof. Lenka Zdeborová and Prof. Florent Krzakala.
- 2022: Post-Doctoral Researcher at the Data Science Department of the International School for Advanced Studies (SISSA) in the group led by Prof. Alessandro Laio.
Teaching Activities
- 2021: Assistant for the course "Machine Learning for Physicists" at École polytechnique fédérale de Lausanne (EPFL).
- 2022-2023: Lecturer in Machine Learning at Medics s.r.l.
- 2023-2024: Tutor in Machine Learning at MIB-Trieste Business School.
Additionally, she has supervised the following master's theses:
- 2021: "Detection of ovarian lesions from ultrasound scans using Deep Learning: from raw data to a ready-to-use medical dataset" by Pio Raffaele Fina, who won the award for best master's thesis of 2021 in computer science at the University of Turin.
- 2023: "A Deep Learning Approach for Segmentation of Ovarian Adnexal masses" by Cecilia Marini.
Research Activity
During her research career, she has collaborated with Riccardo Zecchina, Carlo Baldassi, Carlo Lucibello, Luca Saglietti, Alessandro Ingrosso, Hilbert J. Kappen, and Enzo Tartaglione on characterizing the solution space of non-convex Machine Learning problems, such as the perceptron with discrete weights, and on designing learning algorithms for recurrent neural networks. She subsequently collaborated with Lenka Zdeborová, Florent Krzakala, Marc Mézard, and Bruno Loureiro on advancing techniques from the Physics of Disordered Systems to analyze structured data models, such as the hidden manifold model and Gaussian mixtures, in convex Machine Learning problems. She then worked with Lenka Zdeborová, Andrew Saxe, Luca Saglietti, Stefano Sarao Mannelli, Negar Rostamzadeh, and Alessandro Laio on characterizing transfer learning and data bias towards trustworthy AI. She collaborated with Alessandro Laio, Sebastian Goldt, and Riccardo Rende on theoretical analysis using Statistical Mechanics techniques concerning the functioning of Transformers, the current state-of-the-art among neural network models and a fundamental component of algorithms such as ChatGPT. Currently, she is working with Alessandro Ingrosso, Pietro Rotondo, Rosalba Pacelli, Jean Barbier, Sebastian Goldt, and Alessandro Laio on advancing the theory of transfer learning and self-supervised learning, even in non-convex problems.
She is also a member of the following research projects within the framework of the National Recovery and Resilience Plan (PNRR):
- SEcurity and RIghts in the CyberSpace (SERICS);
- Future Artificial Intelligence Research (FAIR).
She has also contributed with both invited and contributed talks to the following national and international conferences, workshops and schools:
- 2023, Les Houches: Towards a theory of artificial and biological neural networks (contributed);
- 2023, Roma: Interdisciplinary challenges: from non-equilibrium physics to life sciences (invited);
- 2023, Trieste: Youth in High-Dimensions: Recent Progress in Machine Learning, High-Dimensional Statistics and Inference (invited);
- 2023, Parma: XXVII Convegno Nazionale di Fisica Statistica e dei Sistemi Complessi (invited);
- 2023, Trieste: Workshop on Learning and Inference from Structured Data: Universality, Correlations and Beyond (invited);
- 2023, Cargese: Statistical Physics and Machine Learning back together again (contributed);
- 2024, Lausanne: Applied Machine Learning Days (AMLD) EPFL (invited);
- 2024, Trento: Bridging scales: At the crossroads among renormalisation group, multi-scale modelling, and deep learning (invited);
- 2024, Como: Statistical Physics of Deep Learning II (invited).
Institutional Activities and Academic Positions
Since 2024, she has been a Member of the Department Board. Additionally, she is a co-organizer of the international conference Rockin'AI-Roccella Conference on Inference and Learning