Luca Pesce
Center of Mathematical Sciences and Applications, Harvard University .
Hi! I am a postdoctoral fellow at Harvard Center for Mathematical Sciences and Applications (CMSA).
My research seeks to explain why deep learning is so effective in practice. I study how general-purpose training methods exploit the low-dimensional latent structure in high-dimensional data to learn useful representations. Recently I have focused on two questions: continual learning, how models acquire new knowledge without discarding what they already know; and data curation, how to make the most of limited data.
Prior to joining Harvard, I received my PhD from EPFL, where I was advised by Florent Krzakala. During my PhD I spent a few months as an ML research intern at Aqemia, working on generative algorithms for structure-based drug discovery. Earlier, I took part in an International MSc program in Theoretical Physics delocalized in Paris, Trieste, and Turin.
If you want to know more about my research, the list of my publications is here.