1110 対称性を考慮した強化学習の研究(GS-10.16 制御手法)

概要

論文の詳細を見る
This work presents a new approach to reinforcement learning in consideration of symmetries. At first we formalize the concept of symmetry in Markov decision processes (MDPs) and derive theoretical results using this formalism. The second we made a check on these theorem by the simulation. Reinforcement Learning is one of the fields studied briskly recently. But in order to use it in the real world, there are some problems. One of them is associated with very long calculation time and amount of calculation. Then, in order to solve them symmetry is used in this research. At first, the agent check condition of the map. And if the map is symmetry, we use our theorem that is reignforcement of consideration symmetries. When this result is seen, it turns out that it is being early completed by the way of the result in consideration of symmetry.
2004-03-17