Workshop II: PDE and Inverse Problem Methods in Machine Learning - IPAM

Workshop II: PDE and Inverse Problem Methods in Machine Learning

Part of the Long Program High Dimensional Hamilton-Jacobi PDEs
April 20 - 24, 2020

Schedule

All times in this Schedule are Pacific Time (PT)

Monday, April 20, 2020

Morning Session

08:20-08:30
Welcome & Opening Remarks: Dean Miguel García-Garibay (Dean of Physical Sciences, UCLA) and Dima Shlyakhtenko (Director, IPAM)
08:30-08:30
Session Chair Adam Oberman (McGill University)
08:30-09:20
Jianfeng Lu (Duke University Medical Center)
PDE analysis for sampling dynamics and generative models
09:30-09:45
Break
09:45-10:35
Matti Lassas (University of Helsinki)
New deep neural networks solving non-linear inverse problems
10:45-11:00
Break
11:00-11:50
Lightning Poster Presentations: Yiping Lu (Stanford); Tingwei Meng (Brown); Elisa Negrini (WPI); Mirjeta Pasha (Kent State); Lorenz Richter (FU Berlin)

Afternoon Session

12:00-13:00
Lunch (on your own)
13:00-13:50
Michael Jordan (University of California, Berkeley (UC Berkeley))
Optimization with Momentum: Dynamical, Control-Theoretic, and Symplectic Perspectives
14:00-14:15
Break
14:15-15:05
Maxime Laborde (McGill University)
A Lyapunov analysis for accelerated gradient methods: From deterministic to stochastic case

Tuesday, April 21, 2020

Morning Session

08:30-08:30
Session Chair Ivan Yegorov (NDSU)
08:30-09:20
Peng Chen (University of Texas at Austin)
Projected Stein variational methods for high-dimensional Bayesian inversion constrained by large-scale PDEs
09:30-09:45
Break
09:45-10:35
Eldad Haber (University of British Columbia)
A-optimal active learning
10:45-11:00
Break
11:00-11:50
Max Raginsky (University of Illinois at Urbana-Champaign)
A mean-field theory of lazy training in two-layer neural nets: entropic regularization and controlled McKean-Vlasov dynamics

Afternoon Session

12:00-13:00
Lunch (on your own)
13:00-13:50
Stefano Soatto (University of California, Los Angeles (UCLA))
Critical Periods for Regularization in Deep Neural Networks
14:00-14:15
Break
14:15-15:05
Mikhail Belkin (Ohio State University)
Toward a theory of optimization for over-parameterized systems of non-linear equations: the lessons of deep learning

Wednesday, April 22, 2020

Morning Session

08:30-08:30
Session Chair Pratik Chaudhari (University of Pennsylvania)
08:30-09:20
Irène Waldspurger (Université de Paris IX (Paris-Dauphine))
Rank optimality of the Burer-Monteiro factorization
09:30-09:45
Break
09:45-10:35
Yunan Yang (Cornell University)
The joint power of deterministic and Bayesian approaches for seismic inverse problems
10:45-11:00
Break
11:00-11:50
Anthony Yezzi (Georgia Institute of Technology)
Accelerated Optimization in the PDE Framework

Afternoon Session

12:00-13:00
Lunch (on your own)
13:00-13:50
Hossein Mobahi (Google AI)
Differential Operators for Generating Structured Adversarial Examples
14:00-14:15
Break
14:15-15:05
Tom Goldstein (University of Maryland)
An empirical look at generalization in neural nets

Thursday, April 23, 2020

Morning Session

08:30-08:30
Session Chair Dejan Slepcev (Carnegie Mellon University)
08:30-09:20
Wuchen Li (University of California, Los Angeles (UCLA))
Accelerated Information Gradient flow
09:30-09:45
Break
09:45-10:35
Christopher Finlay (McGill University)
Training neural ODEs for density estimation
10:45-11:00
Break
11:00-11:50
Rebecca Willett (University of Chicago)
Learning to Solve Inverse Problems in Imaging

Afternoon Session

12:00-13:00
Lunch (on your own)
13:00-13:50
Quanquan Gu (University of California, Los Angeles (UCLA))
Learning Over-parameterized Neural Networks: From Neural Tangent Kernel to Mean-field Analysis
14:00-14:15
Break
14:15-15:05
Pratik Chaudhari (Amazon)
Learning with few labeled data

Friday, April 24, 2020

Morning Session

08:30-08:30
Session Chair Marc Sedjro (AIMS South Africa)
08:30-09:20
Nicolas Garcia Trillos (University of Wisconsin-Madison)
From clustering with graph cuts to isoperimetric inequalities: quantitative convergence rates of Cheeger cuts on data clouds
09:30-09:45
Break
09:45-10:35
Jeff Calder (University of Minnesota, Twin Cities)
Graph-based semi-supervised learning with very few labels
10:45-11:00
Break
11:00-11:50
Pengchuan Zhang (Microsoft Research)
Training invertible multi-scale deep generative networks for solving high-dimensional Bayesian Inverse Problems

Afternoon Session

12:00-12:00
Conclusion
Printable Version