Department of Computing Science, University of Alberta. Edmonton, Alberta Canada, T6G 2E8. Email: cgao3 AT ualberta.ca Phone: (+1) 780 885 9544
Welcome! I am a PhD student at the Department of Computing Science, co-supervised by Martin Mueller and Ryan Hayward. I received my M.S. degree in 2015 from the Department of Computer Science and Technology from University of Science and Technology of China, where I worked at USTC-Birmingam Joint Research Institute in Intelligent Computation and Its Applications.
My research interests include computer games, reinforcement learning, combinatorial optimization. My current research topic is applying neural networks to the game of Hex.
Code for Unicost SCP Code for MMKP
Chao Gao, Ryan Hayward, Martin Mueller. Move Prediction using Deep Convolutional Neural Networks in Hex. IEEE Transaction on Games, 2017. This paper investigates the move prediction problem in Hex. By learning on MoHex 2.0 self-play generated data, MoHex-CNN achieves 70% winrate against MoHex 2.0 on 13x13 board size after using the learned knowledge as its in-tree prior probability.
Chao Gao, Ryan Hayward, Martin Mueller. Focused depth-first proof number search using convolutional neural networks for the game of Hex. Proceedings of the 26th International Joint Conference on Artificial Intelligence. AAAI Press, 2017. This work uses neural nets to improve proof number search — an algorithm for solving games that can be modeled as AND-OR graph.
Chao Gao, Guanzhou Lu, Xin Yao, Jinlong Li. A Pseudo-gap Enumeration Approach for the Multidimensional Multiple-choice Knapsack Problem. European Journal of Operational Research, 260.1 (2017): 1-11. A family of pseudo-cuts are proposed with a novel concept of pseudo-gap. The derived iterative algorithm becomes the-state-of-art approach for solving MMKPs.
Chao Gao, Xin Yao, Thomas Weise, Jinlong Li. An efficient local search heuristic with row-weighting for the unicost set covering problem. European Journal of Operational Research, 246.3 (2015): 750-761. A simple stochastic local search algorithm is proposed which has similar or even better performance with the best existing algorithm (3-flip search based Lagrangian reduced cost) for solving unicost SCPs.