Departmental Bulletin Paper 意思決定の階層化による強化学習の学習効率の向上
Accuracy Improvement for TSP by Multi-Level Perturbed Parallel Island Model

山森, 一人  ,  渡部, 将人  ,  相川, 勝

45pp.221 - 225 , 2016-07-29 , 宮崎大学工学部
Tracking problem is one of the popular benchmark to evaluate reinforcement learning. In the tracking problem, some hunters trace a target and try to catch target in shorter steps. In the paper, we propose to separate decision marking process of reinforcement learning from two points of view; strategy decision and tactical decision. Strategy decision decides the movement policy of the hunters, and tactical decision decides the movement direction of each hunter. Experimental results showed that our method could catch the target with 54% steps by the conventional reinforcement learning.

Number of accesses :  

Other information