icml paper

UNIST ECE students, (Kyowoon Lee and Sol-A Kim), and Prof. Jaesik Choi’s paper has been accepted at ICML 2018.

Activities May 21, 2018

 

Authors: Kyowoon Lee*, Sol-A Kim * , Jaesik Choi(corresponding author) and Seong-Whan Lee (* contributed equally)

Title: Deep Reinforcement Learning in Continuous Action Spaces: a Case Study in the Game of Simulated Curling

Conference: International Conference on Machine Learning (ICML 2018)

 

A paper written by researchers in the school of ECE in UNIST and Korea university is accepted at ICML. The research team build a new deep learning based reinforcement learning algorithm called Kernel-Regression Deep Reinforcement Learning (KR-UCT). Although existing deep reinforcement learning models (e.g., AlphaGo Lee and AlphaGo Zero) have been successful in discrete action/state spaces, it has not been successful yet to explore continuous actions spaces such as the game of Curling. The model, KR-UCT, suggested by the research team, enables the game playing AI to learn advanced strategies faster than existing deep reinforcement learning methods based on discrete spaces.

 

This is a joint work with Prof. Seong-Whan Lee at Korea University.

ICML is the one of the most prestigious international academic conferences in machine learning.