Kanan Gupta

I am a fourth-year PhD student in the Department of Mathematics at the University of Pittsburgh. My advisor is Dr. Stephan Wojtowytsch. My research interests lie broadly in the mathematics of machine learning. Specifically, I have been focusing on optimization algorithms for machine learning. So far, my work has been on proving accelerated convergence rates for gradient-based first-order algorithms in settings that more closely resemble neural network training, and improving those algorithms in the process. I am also broadly interested in the other areas at the intersection of mathematics and machine learning, like the use of ML for solving PDEs (scientific machine learning) and automated theorem proving.

I started my PhD at Texas A&M University in January 2021 before moving to Pitt with my advisor. Before Texas A&M, I was at Ashoka University where I completed my undergraduate studies with a major in Mathematics and a minor in Computer Science. Afterwards, I did a year-long postgraduate diploma in research at Ashoka University, for which I wrote an expository thesis on K-Theory and C*-Algebras.

Publications

• Nesterov acceleration despite very noisy gradients Kanan Gupta, Jonathan Siegel, and Stephan Wojtowytsch Advances in Neural Information Processing Systems (NeurIPS) 2024. arXiv:2302.05515

• Nesterov acceleration in benignly non-convex landscapes Kanan Gupta and Stephan Wojtowytsch Preprint, under review. arXiv:2410.08395

Service

• Reviewer, ICLR 2025.