Matteo Gamba

I am working as a researcher in Deep Learning as part of Prof. Hossein Azizpour’s group at KTH Royal Institute of Technology. My research is focused on investigating the factors of implicit regularization in supervised and self-supervised learning, as imposed by network architecture, learning algorithm and natural data.

I hold a Doctoral degree in Computer Science from KTH, which I obtained in 2024. My thesis studied implicit regularization of deep networks interpreted as spline functions, focusing on emerging smoothness properties of overparameterized networks.

My work lies at the intersection of geometry and probability, and is aimed at characterizing generalization in deep learning by studying regularity of neural networks interpreted as functions, as well as functional symmetries in parameter space. My research interests include generalization, robustness, mode connectivity, hidden symmetries, and geometry of neural representations.

My long-term vision is making neural networks interpretable and trust-worthy algorithms, to further their use in mission critical and high impact applications.

Teaching

I served as a Teaching Assistant for the following Master-level courses:

• Deep Learning, Advanced Course (2019 - 2024)
• Deep Learning in Data Science (2019 - 2024)
• Artificial Neural Networks and Deep Architectures (2017-2019)
• Computer Security (2016-2019)

Additionally, I was running the Machine Learning reading group at RPL from 2019 to 2022.

System Administration

Between 2019 and 2024 I have been in charge of procuring and administering a SLURM-based GPU cluster with a hundred GPUs, providing compute to the RPL division (100+ users).

Other

I am a founding member of the GNU Education project, which promotes free software in education.

I am part of the NGO Hipatia and co-founder of the project GURI, which promote the freedom of sharing knowledge and foster life-long learning through free software.