Opening Workshop: August 12-16, 2019

Location

This workshop took place at Gross Hall on the campus of Duke University.

Description

The SAMSI Deep Learning program brought together mathematical, statistical, and computer scientists interested in understanding the theoretical capabilities and limitations of deep learning methodology. This workshop featured overview talks that introduce the basic concepts of deep networks (including convolutional neural networks, recursive neural networks, generative adversarial networks, and various kinds of autoencoders). Additionally, there were a series of talks that highlighted recent research in this area. Later in the week participants broke out into working groups, and developed plans for their semester of work on these topics. On the final day, the leaders of the working groups shared their plans of their group with all of the workshop participants.

Confirmed speakers for this event include:

Schedule and Supporting Media

Printed Schedule
Titles & Abstracts
Participants List

Monday, August 12, 2019
Ahmadieh Family Auditorium, Gross Hall, Rm 107, Duke University, Durham, NC

Description Speaker Slides Video
Registration
Welcome David Banks, Duke University and Director, SAMSI
A Survey of Statistical Research in Neural Networks and Deep Learning Xiaoming Huo, Georgia Institute of Technology
Admissibility of Solution Estimators for Stochastic Optimization Amitabh Basu, Johns Hopkins University
Statistical and Computational Guarantees of EM with Random Initialization Harrison Zhou, Yale University
Robust Information Bottleneck Poh-Ling Loh, University of Wisconsin
Towards Deep Learning: Understanding Statistical Properties by Bridging Convex and Nonconvex Optimization Jianqing Fan, Princeton University
Horseshoe Regularization for Machine Learning in Complex and Deep Models Anindya Bhadra, Purdue University
Posterior Concentration for Sparse Deep Learning Veronika Rockova, University of Chicago
Deep Compositional Spatial Models Andrew Zammit Mangion, University of Woolongong, Australia
Training DNN with Dynamic SMD Shih-Kang Chao, University of Missouri
Posters and Reception

Tuesday, August 13, 2019
Ahmadieh Family Auditorium, Gross Hall, Rm 107, Duke University, Durham, NC

Description Speaker Slides
On Adversarial Learning Larry Carin, Duke University
Deep Instrumental Variables Estimator Ruiqi Liu, University of Indiana
Improving Generative Models Junier Oliva, University of North Carolina at Chapel Hill
Learning to Solve Inverse Problems in Imaging Rebecca Willett, University of Chicago
An Adaptively Weighted Stochastic Gradient MCMC Algorithm for Global Optimization in Deep Learning Faming Liang, Purdue University
Information Geometric and Topological Approaches to Deep Learning Wyatt Bridgman and Sorin Mitran, University of North Carolina at Chapel Hill
Domain Adaptation Challenges in Genomics: a deep learning take on medical pathology Bianca Dumitrascu, Princeton University and SAMSI
Neural Network Density Estimation Deborshee Sen, Duke University and SAMSI
Complexity Bounds for Deep Learning Networks via the Probabilistic Method Jason Klusowski, Rutgers University

Wednesday, August 14, 2019
Ahmadieh Family Auditorium, Gross Hall, Rm 107, Duke University, Durham, NC

Description Speaker Slides
Statistical Inference for Online Decision Making via Stochastic Gradient Descent Rui Song, N.C. State University
Modern Statistical Theory Inspired by Deep Learning Guang Cheng, Purdue University
Deep ReLU Networks Viewed as a Statistical Method Johannes Schmidt-Hieber, University of Twente
ReLU regression: Complexity and Approximation Algorithms Yao Xie, Georgia Institute of Technology
ProxSARAH Algorithms for Stochastic Composite Nonconvex Optimization Quoc Tran-Dinh, University of North Carolina at Chapel Hill
Deep Models for Improved Topic Recovery Deanna Needell, UCLA
Group-equivariant Representation by Jointly Decomposed Convolution Xiuyuan Cheng, Duke University
Optimization and Learning with Nonconvex Functional Constraints Guanghui (George) Lan, Georgia Institute of Technology
Working Group Formation

Thursday, August 15, 2019
Breakout Rooms: 318, 324, 351, Gross Hall

Description Location
Working Group Planning Meeting: Bayesian Methods in Deep Learning Gross Hall, Rm 318
Working Group Planning Meeting: Teaching Deep Learning Gross Hall, Rm 324
Working Group Planning Meeting: Uncertainty Quantification for Deep Models Gross Hall, Rm 351
Working Group Planning Meeting: Novel Applications Gross Hall, Rm 318
Working Group Planning Meeting: Interpretable Deep Networks Gross Hall, Rm 324
Working Group Planning Meeting: Architecture Search Gross Hall, Rm 351
Working Group Planning Meeting: Deep Learning for Dimension Reduction Gross Hall, Rm 318
Working Group Planning Meeting: Statistical Optimality Gross Hall, Rm 324
Working Group Planning Meeting: Regularization Techniques for Training Deep Networks Gross Hall, Rm 351

Friday, August 16, 2019
Ahmadieh Family Auditorium Gross Hall, Rm 107, Duke University, Durham, NC

Description
Working Group Leaders provide description of research plans
Working Group Leaders provide description of research plans
Closing Remarks

Questions: email dl@samsi.info