Alex Damian

Alex Damian is an Assistant Professor at MIT in Mathematics and EECS. His research focuses on the mathematical foundations of deep learning, with an emphasis on optimization dynamics and representation learning. He was previously a research fellow at the Kempner Institute at Harvard University. He received his Ph.D. in Applied and Computational Mathematics from Princeton University, where he was advised by Jason D. Lee, and his B.S. in Mathematics from Duke University.

AI2050 Project

Training an AI model requires following a complex recipe involving dozens of settings, including learning rates, initializations, and normalization schemes, which have been carefully tuned through years of trial and error. These recipes are also brittle and often fail to adapt to new architectures or datasets, slowing progress in AI. Damian’s project develops a mathematical theory of how optimizers and neural network architectures interact, and then translates it into algorithms that automatically derive effective training recipes from a model’s architecture. The goal is to enable new applications of AI by making it easier and faster to train new models without excessive trial and error.