Abstract
Conference spotlight on DeepReDuce, a set of optimizations for selectively reducing ReLU operations to lower the cost of cryptographically secure private inference while preserving model quality.
Date
Jul 1, 2021
Location
Virtual Conference
ICML 2021 spotlight talk on criticality-based ReLU reduction for fast private inference.
Nandan Kumar Jha
Ph.D., New York University · Representation Learning, Scaling Laws, and High-Dimensional Learning Dynamics
I study nonlinear representation dynamics in large language models, focusing on how nonlinearities, architecture, and optimization jointly shape representational geometry, scaling behavior, and usable computational capacity.