Schedule
Tuesday, April 5
| Time (ET) | |
| 10:30am-10:40 | Opening remarks |
| 10:40-11:40 | Invited talk Lenka Zdeborova, EPFL Overparametrization: Insights from Solvable Models Abstract |
| 11:40-12:05 | Contributed talk Bruno Loureiro, EPFL Fluctuations, Bias, Variance & Ensemble of Learners: Exact Asymptotics for Convex Losses in High-Dimension Abstract |
| 12:05-12:15 | Break |
| 12:15-1:15 | Invited talk Andrea Montanari, Stanford From Projection Pursuit to Interpolation Thresholds in Small Neural Networks Abstract |
| 1:15-1:40 | Contributed talk Pratik Patil, Carnegie Mellon University Revisiting Model Complexity in the Wake of Overparameterized Learning Abstract |
| 1:40-2:00 | Break |
| 2:00-3:00 | Invited talk Jeffrey Pennington, Google Research Covariate Shift in High-Dimensional Random Feature Regression Abstract |
| 3:00-4:00 | Lightning talk session #1 |
| 4:00-4:05 | Break |
| 4:05-5:05 | Invited talk Vidya Muthukumar, Georgia Tech Classification versus Regression in Overparameterized Regimes: Does the Loss Function Matter? Abstract |
| 5:05-5:30 | Contributed talk Spencer Frei, UC Berkeley Benign Overfitting without Linearity: Neural Network Classifiers Trained by Gradient Descent for Noisy Linear Data Abstract |
| 5:30-6:30 | Lightning talk session #2 |
Wednesday, April 6
| Time (ET) | |
| 10:30am-11:30 | Invited talk Francis Bach, Ecole Normale Supérieure The Quest for Adaptivity Abstract |
| 11:30-11:55 | Contributed talk Mariia Seleznova, LMU Munich Neural Tangent Kernel Beyond the Infinite-Width Limit: Effects of Depth and Initialization Abstract |
| 11:55-12:10 | Break |
| 12:10-1:10 | Invited talk Daniel Hsu, Columbia University Computational Lower Bounds for Tensor PCA Abstract |
| 1:10-1:35 | Contributed talk Nikhil Ghosh, UC Berkeley The Three Stages of Learning Dynamics in High-dimensional Kernel Methods Abstract |
| 1:35-1:45 | Break |
| 1:45-2:45 | Lightning talk session #3 |
| 2:45-3:10 | Contributed talk Lorenzo Luzi, Rice University Double Descent and Other Interpolation Phenomena in GANs Abstract |
| 3:10-4:10 | Invited talk Caroline Uhler, MIT Update: The talk will be given by Adityanarayanan Radhakrishnan, MIT Over-parameterized Autoencoders and Causal Transportability Abstract |
| 4:10-4:20 | Break |
| 4:20-5:20 | Invited talk Edgar Dobriban, University of Pennsylvania T-Cal: An Optimal Test for the Calibration of Predictive Models Abstract |
| 5:20-6:25 | Lightning talk session #4 |
| 6:25-6:30 | Closing remarks |
Lightning Talks
Session #1: Tuesday, April 5, 3:00pm-4:00pm (ET)
On the Double Descent of Random Features Models Trained with SGD
Fanghui Liu (EPFL); Johan Suykens (KU Leuven); Volkan Cevher (EPFL)
Phase diagram of Stochastic Gradient Descent in High-Dimensional Two-Layer Neural Networks
Rodrigo Veiga (EPFL); Ludovic Stephan (EPFL); Bruno Loureiro (EPFL); Florent Krzakala (EPFL); Lenka Zdeborova (EPFL)
Precise Asymptotic Analysis for Double Descent under Generic Convex Regularization
David Bosch (Chalmers University); Ashkan Panahi (Chalmers University); Ayca Ozcelikkale (Uppsala University); Devdatt Dubhashi (Chalmers University)
Investigating Reproducibility and Double Descent from the Decision Boundary Perspective
Gowthami Somepalli (University of Maryland); Liam Fowl (University of Maryland); Arpit Bansal (University of Maryland); Ping-yeh Chiang (University of Maryland); Yehuda Dar (Rice University); Richard Baraniuk (Rice University); Micah Goldblum (NYU); Tom Goldstein (University of Maryland)
Overfitting in Transformers – The Slingshot Mechanism
Vimal Thilak (Apple); Etai Littwin (Apple); Shuangfei Zhai (Apple); Omid Saremi (Apple); Joshua M Susskind (Apple)
Session #2: Tuesday, April 5, 5:30pm-6:30pm (ET)
Benign Overfitting in Multiclass Classification: All Roads Lead to Interpolation
Ke Wang (UC Santa Barbara); Vidya Muthukumar (Georgia Tech); Christos Thrampoulidis (University of British Columbia)
Over-parameterization: A Necessary Condition for Models that Extrapolate
Roozbeh Yousefzadeh (Yale University)
Random Feature Amplification: Feature Learning and Generalization in Neural Networks
Spencer Frei (UC Berkeley); Niladri S Chatterji (UC Berkeley); Peter Bartlett (UC Berkeley)
Benign Overfitting in Overparameterized Time Series Models
Shogo Nakakita (The University of Tokyo); Masaaki Imaizumi (The University of Tokyo / RIKEN AIP)
Benign Overfitting in Conditional Average Treatment Effect Prediction with Linear Regression
Masahiro Kato (Cyberagent / The University of Tokyo); Masaaki Imaizumi (The University of Tokyo)
Session #3: Wednesday, April 6, 1:45pm-2:45pm (ET)
Benign Overfitting in Two-layer Convolutional Neural Networks
Yuan Cao (The University of Hong Kong); Zixiang Chen (UCLA); Mikhail Belkin (UC San Diego); Quanquan Gu (UCLA)
Error Rates for Kernel Methods under Source and Capacity Conditions
Hugo CUI (EPFL); Bruno Loureiro (EPFL); Florent Krzakala (EPFL); Lenka Zdeborova (EPFL)
Effective Number of Parameters in Neural Networks via Hessian Rank
Sidak Pal Singh (ETH Zurich); Gregor Bachmann (ETH Zurich); Thomas Hofmann (ETH Zurich)
Locality Defeats the Curse of Dimensionality in Convolutional Teacher-Student Scenarios
Alessandro Favero (EPFL); Francesco Cagnetta (EPFL); Matthieu Wyart (EPFL)
Relative Stability Toward Diffeomorphisms Indicates Performance in Deep Nets
Leonardo Petrini (EPFL); Alessandro Favero (EPFL); Mario Geiger (EPFL); Matthieu Wyart (EPFL)
Session #4: Wednesday, April 6, 5:20pm-6:25pm (ET)
On How to Avoid Exacerbating Spurious Correlations When Models are Overparameterized
Tina Behnia (University of British Columbia); Ke Wang (UC Santa Barbara); Christos Thrampoulidis (University of British Columbia)
Consistent Interpolating Ensembles
Yutong Wang (University of Michigan); Clayton Scott (University of Michigan)
Provable Boolean Interaction Recovery from Tree Ensemble obtained via Random Forests
Merle Behr (UC Berkeley); Yu Wang (UC Berkeley); Xiao Li (UC Berkeley); Bin Yu (UC Berkeley)
Mitigating Multiple Descents: Model-Agnostic Risk Monotonization in High-Dimensional Learning
Pratik Patil (Carnegie Mellon University); Arun Kuchibhotla (Carnegie Mellon University); Alessandro Rinaldo (Carnegie Mellon University); Yuting Wei (University of Pennsylvania)
On the Implicit Bias Towards Minimal Depth of Deep Neural Networks
Tomer Galanti (MIT)