Positions:
AWS AI Labs, Applied Scientist Intern, California (Jul-Oct 2024)
PhD in Machine Learning, EPFL (2020-Feb 25)
During my research experience, I used a variety of Machine Learning approaches, including Large Language Models, Diffusion models, Deep Learning and Kernel methods. The outcome of my research is summarized in the following papers and presentations.
Papers:
(2024) How Deep Networks Learn Sparse and Hierarchical Data: the Sparse Random Hierarchy Model, Tomasini U.M. and Wyart M., ICML 2024 (Spotlight).
(2023) How Deep Neural Networks Learn Compositional Data: The Random Hierarchy Model, Petrini L., Cagnetta F., Tomasini U.M., Favero A. and Wyart M. , Physical Review X.
(2022) How deep convolutional neural networks lose spatial information with training, Tomasini U.M., Petrini L., Cagnetta F. and Wyart M., Machine Learning: Science and Technology (2023).
(2022) Failure and success of the spectral bias prediction for Laplace Kernel Ridge Regression: the case of low-dimensional data, Tomasini U.M., Sclocchi A. and Wyart M., ICML 2022 (Spotlight).
(2021) Predictors and Predictands of Linear Response in Spatially Extended Systems, Tomasini U.M. and Lucarini V., Eur. Phys. J. Spec. Top. (2021).
Outreach:
(2023) Interview: Umberto Tomasini: da Mirandola alla Svizzera, dalla Fisica alla teoria dell’Intelligenza Artificiale (Italian).
(2023) Talk: Statistical Physics & Machine Learning back together again, Cargese.
(2023) Poster: Princeton Machine Learning Theory Summer School, Princeton.
Awarded: Princeton ML Theory Summer School travel grant
(2023) Talk: Physically-informed perturbations and their applications in interpretable AI, Hes-So Valais.
(2023) Poster: Physics4ML workshop, Kigali ICLR 2023.
(2022) Poster: Statistical Physics & Machine learning summer school, Les Houches.
(2022) Poster: AI For Science first workshop, EPFL.
(2022) Talk: Physics of Data first workshop, Venice.
(2022) Poster: Simons Collaboration on Cracking the Glass Problem Annual Meeting, New York.
(2021-22) Talks: SPOC+IdePHICS+PCSL seminar (Profs. Zdeborová, Krzakala, Wyart’s groups), EPFL.
(2021) Academic Visit: Collaboration on unsupervised dimensional reduction, SISSA, Prof. Laio.
(2021) Talk: University of Reading SIAM-IMA Student chapter, Reading.