Schedule

Monday - January 14

8:30 - 9:00: - Priors for deep infinite networks - Sam Schoenholz

9:00 - 9:30: - Fluctuation-dissipation relation for stochastic gradient descent - Sho Yaida

9:30 - 9:45: - Break

9:45 - 10:15: - Over-parametrization in neural networks: observations and a definition - Levent Sagun

10:15 - 10:45: - Generalization under noise, adversarial examples, and data augmentation - Ekin Cubuk

10:45 - 4:15: - Ski Break

4:15 - 4:45 - Neuroscience-based machine learning - Dmitri Chklovskii

4:45 - 5:15 - Unsupervised learning on a sphere - Dmitry Krotov

5:15 - 5:30 - Break

5:30 - 6:00 - Neural network classifiers on NISQ-regime processors - Jack Hidary & Stefan Leichenauer

6:00 - 6:30 - Signal propagation and gradient stability in the LSTM and GRU - Dar Gilboa

6:30 - 6:45 - Break

6:45 - 7:15 - Essentially no barriers in neural network energy landscape - Felix Draxler

7:15 - 7:45 - Gradient and hessian properties in logistic regression - Vladimir Kirilin

Tuesday - January 15

8:30 - 9:00: - Scalable natural gradient training of neural networks - Roger Grosse

9:00 - 9:30: - The role of a layer in supervised learning - Zohar Ringel

9:30 - 9:45: - Break

9:45 - 10:15: - Statistics of random neural networks - Jeffrey Pennington

10:15 - 10:45: - An empirical model of large-batch training - Sam McCandlish

10:45 - 4:15: - Ski Break

4:15 - 4:45 - Visualizing Probabilities: intensive principal component analysis - Katherine Quinn

4:45 - 5:15 - On the spectral bias of neural networks - Aristide Baratin

5:15 - 5:30 - Break

5:30 - 6:00 - Noise2Self: blind de-noising with self-supervision - Joshua Batson

6:00 - 6:30 - Geometric deep learning in a non-commutative world - Joan Bruna

6:30 - 6:45 - Break

6:45 - 7:15 - TherML: a thermodynamics of machine learning - Alexander Alemi

7:15 - 7:45 - High-dimensional manifold classification by multilayered perceptrons -Jonathan Kadmon

Wednesday - January 16

8:30 - 9:00: - Replica symmetry breaking in bipartite spin glasses and neural networks - Gavin Hartnett

9:00 - 9:30: - Wide, deep neural networks as gaussian processes - Yasaman Bahri

9:30 - 9:45: - Break

9:45 - 10:15: - The role of over parametrization in the generalization of neural networks - Behnam Neyshabur

10:15 - 10:45: - What can stochastic differential equations tell us about SGD and natural gradient descent? - Samuel Smith

10:45 - 4:15: - Ski Break

4:30 - 5:30: - Physics Café - Yasaman Bahri & Dmitry Krotov (Wheeler Opera House)

5:30 - 6:30: - Public Lecture (Rise of the machines: deep learning from backgammon to Skynet) - Paul Ginsparg

(Wheeler Opera House)

Thursday - January 17

8:30 - 9:00: - EL2O and MPM: alternatives to variational inference and MAP/MLE - Uros Seljak

9:00 - 9:30: - A mean field view of the landscape of two layers neural networks - Andrea Montanari

9:30 - 9:45: - Break

9:45 - 10:45: - Panel discussion - A. Alemi, B. Neyshabur, X. Qi, E. Silverstein, J. Sohl-Dickstein

10:45 - 4:15: - Ski Break

4:15 - 4:45 - Introduction to topological data analysis - Maxime Gabella

4:45 - 5:15 - Local deformations and global features in persistent homology, machine learning, and physics - Alex Cole

5:15 - 5:30 - Break

5:30 - 6:00 - Collision Course: particle physics as a machine learning testbed - Jesse Thaler

6:00 - 6:30 - ML for sub-grid physics - David Spergel

6:30 - 6:45 - Break

6:45 - 7:15 - Parameter estimation in intractable probabilistic models by minimum probability flow learning - Jascha Sohl-Dickstein

7:15 - 7:45 - The Goldilocks zone: Understanding neural network loss landscapes - Adam Scherlis

8:15 - 10:15 - Banquet

Friday - January 18

8:30 - 9:00: - JUNIPR: a framework for unsupervised machine learning in particle physics - Anders Andreassen

9:00 - 9:30: - Searching for new physics with deep autoencoders - David Shih

9:30 - 9:45: - Break

9:45 - 10:15: - Why ConvNets generalize better than fully-connected networks? - Lechao Xiao

10:15 - 10:45: - Impossibility and uncertainty theorems in AI value alignment - Peter Eckersley

10:45 - 4:15: - Ski Break

4:15 - 4:45 - Approximate optimization and inference with tensor networks - Masoud Mohseni

4:45 - 5:15 - Machine Learning by a String (Theorist) - Michael Douglas

5:15 - 5:30 - Break

5:30 - 6:00 - Optimal SGD hyperparameters for fully connected networks - Daniel Park

6:00 - 6:30 - Diversity versus depth in neural representations - Samuel Ocko

6:30 - 6:45 - Break

6:45 - 7:15 - "Wavefunctions" of data: applying tensor network methods from physics to machine learning - Miles Stoudenmire