<< Wednesday, May 01, 2019 >>

Wednesday, May 1, 2019

Scientific Computing and Matrix Computations Seminar: Why Deep Learning Works: Traditional and Heavy-Tailed Implicit Self-Regularization in Deep Neural Networks

Seminar: Scientific Computing: CS | May 1 | 12-1 p.m. | 380 Soda Hall

 Michael W. Mahoney, ICSI and Department of Statistics, University of California at Berkeley

 Electrical Engineering and Computer Sciences (EECS)

Random Matrix Theory (RMT) is applied to analyze the weight matrices of Deep Neural Networks (DNNs), including both production quality, pre-trained models and smaller models trained from scratch. Empirical and theoretical results clearly indicate that the DNN training process itself implicitly implements a form of self-regularization, implicitly sculpting a more regularized energy or penalty...   More >

Dissertation Talk: Sample Complexity Bounds for the Linear Quadratic Regulator

Seminar: Dissertation Talk: EE | May 1 | 3-4 p.m. | 380 Soda Hall

 Stephen Tu, UC Berkeley

 Electrical Engineering and Computer Sciences (EECS)

Reinforcement learning (RL) has demonstrated impressive performance in various domains such as video games, Go, robotic locomotion, and manipulation tasks. As we turn towards RL to power autonomous systems in the physical world, a natural question to ask is, how do we ensure that the behavior observed in the laboratory reflects the behavior that occurs when systems are deployed in the real world?...   More >

Rapidly mixing random walks on matroids and related objectsidly mixing random walks on matroids and related objects

Seminar: CS | May 1 | 3-4 p.m. | 1011 Evans Hall

 Nima Anari, Stanford University

 Department of Statistics

A central question in randomized algorithm design is what kind of distributions can we sample from efficiently? On the continuous side, uniform distributions over convex sets and more generally log-concave distributions constitute the main tractable class. We will build a parallel theory on the discrete side, that yields tractability for a large class of discrete distributions. We will use this...   More >

Center for Computational Biology Seminar: Sohini Ramachandran, Associate Professor, Brown University

Seminar: Biosystems and Computational Biology: CS | May 1 | 4:30-5:30 p.m. | 125 Li Ka Shing Center

 Center for Computational Biology

Leveraging linkage disequilibrium to identify adaptive and disease-causing mutations

Correlation among genotypes in human population-genetic datasets complicates the localization of both adaptive mutations and disease-causing mutations. I will describe our latest efforts to develop new methods for localizing adaptive and disease-causing mutations, motivated by (1) incorporating...   More >