[22p-E102-1~14] FS.1 Focused Session "AI Electronics"

△ [22p-E102-10] Collective decision making through quantum interference and its application to multi-agent reinforcement learning

〇(M1)Hiroaki Shinkawa¹, Nicolas Chauvet¹, Guillaume Bachelier², Serge Huant², Andre Roehm¹, Ryoichi Horisaki¹, Makoto Naruse¹ (1.Univ. Tokyo, 2.Univ. Grenoble Alpes)

Keywords:quantum interference, reinforcement learning, bandit problem

Previous research realized a conflict-free collective decision-making system using orbital angular momentum of photons through quantum interference to solve the competitive multi-armed bandit problem. This research extended the system to make collective decision-making in more general multi-agent reinforcement learning where the state changes due to the action. We confirmed that the speed of the learning increases if the agents make decisions based on the bandit algorithm instead of making random decisions and if they make conflict-free decisions.