Presenter: Kazuki Yoshizoe
RIKEN Center for Advanced Intelligence Project
kazuki.yoshizoe@riken.jp

Machine Learning and Search are two major domains in Artificial Intelligence. Recent success of the AlphaGo is based on a combination of Deep Learning and Monte Carlo Tree Search. This talk focuses on the challenges in parallelizing search algorithms and deep neural network training. Although search algorithms are useful tools, parallelization is often considered to be difficult. In this talk, I present the parallel algorithms for Depth-First Search, A* search, and Monte-Carlo Tree Search. Additionally, I plan to introduce recent results and problems in parallelizing deep neural network training.

Slides

Kazuki Yoshizoe received his PhD from the University of Tokyo based on his work in AND-OR tree search algorithms. When he was a PhD candidate, he studied at a parallel computing laboratory. Then, he got interested in AI, especially search algorithm, including Monte-Carlo Tree Search which is famous for the success in computer Go. For recent several years, his main interest was in scalable parallel search algorithms. From 2017, he had joined RIKEN Center for Advanced Intelligence Project (AIP).

Last modified: Thursday, June 29, 2017, 6:00 PM