–
Opening Remarks
–
Benign Overfitting and Double Descent

Peter Bartlett (UC Berkeley)
–
Toward a Theory of Optimization for Over-Parameterized Systems of Non-Linear Equations: The Lessons of Deep Learning

Misha Belkin (UC San Diego)
–
Overparameterization in Regression vs Classification Problems

Anant Sahai (UC Berkeley)
–
Two Vignettes on Narrow Networks

Matus Telgarsky (University of Illinois)
–
Finite vs. Infinite Neural Networks

Jascha Sohl-Dickstein (Google Brain)
–
Gather.town
–
Break
–
The Large Learning Rate Phase of Deep Learning

Yasaman Bahri (Google Brain)
–
Finite-Sample Analysis of Interpolating Linear Classifiers in the Overparameterized Regime

Phil Long (Google)
–
Chasing the Long Tail: What Neural Networks Memorize and Why

Vitaly Feldman (Apple)
–
On Uniform Convergence and Interpolation Learning

Nati Srebo (Toyota Technical Institute at Chicago)
–
Approximation Schemes for ReLU Regression

Surbhi Goel (Microsoft Research)
–
Gather.town
–
Reverse Isoperimetric Inequalities for Adversarial Machine Learning

Varun Jog (University of Wisconsin, Madison)
–
Precise Tradeoffs in Adversarial Training

Mahdi Soltanolkotabi (University of Southern California)
–
Extracting Robust and Accurate Features via a Robust Information Bottleneck

Po-Ling Loh (University of Wisconsin, Madison)
–
Neural Perceptual Adversarial Robustness: Generalizable Defenses to Unforeseen Attacks

Soheil Feizi (University of Maryland)
–
A Generalized Neural Tangent Kernel Analysis for Two-layer Neural Networks

Quanquan Gu (UCLA)
–
Gather.town
–
Break
–
The Implicit Regularization of Stochastic Gradient Flow for Least Squares

Edgar Dobriban (University of Pennsylvania)
–
The Complexity of Non-Convex Stochastic Optimization

Dylan Foster (MIT)
–
Initialization Scale vs Training Accuracy Effects on Implicit Bias in Deep Linear Classification

Suriya Gunasekar (Microsoft Research)
–
Analytic Characterization of the Hessian in Shallow ReLU Models: A Tale of Symmetry

Yossi Arjevani (New York University)
–
Beyond Linearization: On Quadratic and Higher-order Taylor Models of Wide Neural Networks

Yu Bai (Salesforce Research)
–
Gather.town
–
A Trichotomy of Rates in Supervised Learning

Amir Yehudayoff (Technion)
–
Contrastive Estimation Reveals Topic Posterior Information to Linear Models

Daniel Hsu (Columbia University)
–
Beyond Signal Propagation: Is Feature Diversity Necessary in Deep Neural Network Initialization?

Daniel Soudry (Technion)
–
Few-Shot Learning via Learning the Representation, Provably

Qi Lei (Princeton University)
–
How to Make Our Models Learn (the Right Things)?

Aleksander Madry (MIT)
–
Gather.town
–
Break
–
Generalization via Derandomization With an Application to Interpolating Predictors

Gintare Karolina Dziugaite (Element AI)
–
How Does Our Mind Store Information in Memory?

Rina Panigrahy (Google Research)
–
Fine-Tuning Language Models With Human Feedback

Paul,
Christian
–
Equilibrium in Nonconvex-Nonconcave Min-Max Optimization and GANs

Nisheeth Vishnoi (Yale)
–
Anatomy of Catastrophic Forgetting: Hidden Representations and Task Semantics

Maithra Raghu (Cornell University)
–
Data Augmentation as Stochastic Optimization

Boris Hanin (Princeton)
–
Gather.town
–
Tight Bounds on Minimax Regret under Logarithmic Loss via Self-Concordance

Daniel Roy (University of Toronto)
–
The Intriguing Role of Module Criticality in the Generalization of Deep Networks

Hanie Sedghi (Google Brain)
–
Predicting What You Already Know Helps: Provable Self-Supervised Learning

Jason Lee (Princeton University)
–
Probabilistic Approximate Variants of Dimensional and Margin Complexity

Pritish Kamath (Toyota Technological Institute at Chicago)
–
Shape Matters: Understanding the Implicit Bias of the Noise Covariance

Tengyu Ma (Stanford University)
–
Gather.town
–
Break
–
Group Discussion