The geometry of deep learning
My research aims at understanding the essence of deep learning and what its limits might be. I approach this problem from a background in geometry, and seek to contribute to the construction of geometric foundations for optimisation, architecture and statistical learning that bring our theory of deep learning closer to what we see in practice. You can find a (hopefully accessible) presentation on my recent work here.
I also maintain an interest in pure mathematics and am troubled by the implications of the recent application of AI to mathematics. Ultimately, however, I believe that if we maintain a commitment to human insight then AI and its theoretical study have the potential to develop a relationship with mathematics analogous to that enjoyed by mathematical physics: one of mutual enrichment and benefit.