All talks listed in Pacific Time. Schedule subject to change.

–
Coffee and Check-In
–
Opening Remarks
–
Transformers can learn compositional function

Jason Lee (Princeton University)
–
Talk By

Binghui Peng (University of Maryland)
–
Learning from the Right Teacher in Knowledge Distillation

Bingbin Liu (Carnegie Mellon University)
–
Talk By

Daniel Hsu (Columbia University)
–
Lunch (on your own)
–
Data Uniformity Improves Training Efficiency and More, with a Convergence Framework Beyond the NTK Regime

Yuqing Wang (Johns Hopkins University)
–
Training Dynamics of Softmax Self-Attention: Global Convergence and Neural Scaling Laws

Gautam Goel (Simons Institute, UC Berkeley)
–
Talk By

Ryan Tibshirani (UC Berkeley)
–
Break
–
Panel
–
Reception
–
Coffee and Check-In
–
Deep sequence models tend to memorize geometrically; it is unclear why.

Vaishnavh Nagarajan (Google)
–
Talk By

Vatsal Sharan (University of Southern California)
–
Talk By

Mahdi Soltanolkotabi (University of Southern California)
–
Talk By

Andrej Riteski (Carnegie Mellon University)
–
Lunch (on your own)
–
Talk By

Nikita Zhivotovskiy (UC Berkeley)
–
Talk By

Ankur Moitra (MIT)
–
Talk By

Mikhail Belkin (UCSD)
–
Break
–
Panel
–
Coffee and Check-In
–
Learning AC^0 Circuits Beyond Product Distributions

Arsen Vasilyan (Simons Institute)
–
Neural Algorithic Design

Yusu Wang (UCSD)
–
Gradient Descent Dominates Ridge: A Statistical View on Implicit Regularization

Jingfeng Wu (UC Berkeley)
–
Generalizing Beyond, and Without a Population Distributions

Nathan Srerbo (Toyota Technological Institute at Chicago)
–
Lunch (on your own)
–
Talk By

Surbhi Goel (University of Pennsylvania)
–
Distortion of Learning from Human Preferences

Nika Haghtalab (UC Berkeley)
–
Break
–
Discussion
–
Coffee and Check-In
–
Talk by

Matus Telgarsky (NYU)
–
Parallelizing autoregression and diffusion models

Thuy-Duong (June) Vuong (UC San Diego)
–
When do spectral gradient updates help in deep learning?

Damek Davis (University of Pennsylvania)