Abstract
This talk presented entropy as a useful lens for understanding and controlling private language-model inference, connecting attention entropy, nonlinear operation costs, and efficient Transformer design under cryptographic constraints.
Date
Apr 1, 2025 2:00 PM — 3:00 PM
Location
New York University
New York, NY
Invited seminar in the CILVR Seminar Series at NYU Center for Data Science, discussing entropy dynamics and efficient private language-model inference.
Nandan Kumar Jha
Ph.D., Electrical and Computer Engineering
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.