ECE Seminar Lecture Series
Towards "Ideal" Model Training on Non-ideal Analog Computing Hardware
Tianyi Chen, Associate Professor of Electrical and Computer Engineering, Cornell University
Wednesday, April 1, 2026
Noon–1 p.m.
601 Computer Studies Building
Abstract: Large-scale AI models are increasingly limited by the energy inefficiency of conventional digital processors. Analog in-memory computing offers a promising alternative by performing AI computations directly in hardware, potentially achieving one to two orders of magnitude improvement in energy efficiency. However, training neural networks directly on non-ideal analog devices remains largely unexplored. Recent empirical evidence shows that the standard workhorse of training - stochastic gradient descent (SGD) - can fail to converge when deployed on hardware exhibiting asymmetric updates, read noise, and limited precision.
In this talk, we develop a mathematical framework to characterize training dynamics on imperfect analog devices. Our analysis reveals how device physics induces structured gradient bias and reshapes the effective optimization objective, explaining the non-convergence behavior of vanilla SGD and the emergence of asymptotic training error. Building on this theory, we introduce grounded algorithmic correction mechanisms that compensate for update bias and restore convergence guarantees. We will conclude the talk by showing encouraging simulations and pointing out future directions in this area.
Bio: Tianyi Chen an associate professor of electrical and computer engineering at Cornell Tech and Cornell Engineering. Dr. Chen's His research focuses on the theoretical foundations of bilevel and multi-objective optimization, with recent applications to efficient AI inference and training through analog computing. His work bridges theory and practice, contributing to multiple IBM’s products and patents.
Dr. Chen is the inaugural recipient of IEEE Signal Processing Society (SPS) Best PhD Dissertation Award in 2020, a recipient of NSF CAREER Award in 2021, and several industrial research awards including Amazon Research Award and Cisco Research Award. He is also the recipient of several best (student) paper awards including the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) in 2021, and the IEEE SPS Young Author Best Paper Award in 2024.