ASN: action semantics network for multiagent reinforcement learning

Article Computer Science, Artificial Intelligence

UNMAS: Multiagent Reinforcement Learning for Unshaped Cooperative Scenarios

Jiajun Chai et al.

Summary: This article introduces a new method called UNMAS, which adapts to changes in the number and size of multiagent systems. By using self-weighting mixing networks and individual action-value networks, UNMAS outperforms other MARL algorithms in StarCraft II micromanagement scenarios, especially in the most difficult scenario.

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (2023)

Add to Collection

Article Computer Science, Information Systems

A Novel and Efficient Influence-Seeking Exploration in Deep Multiagent Reinforcement Learning

Byunghyun Yoo et al.

Summary: This paper proposes a novel exploration method, ISE, which speeds up the learning process by quantifying the influences between agents. By using the variance of action values to measure the influences of other agents and applying it to the exploration scheme, the proposed method achieved significant performance improvements in different tasks.

IEEE ACCESS (2022)

Add to Collection

Article Multidisciplinary Sciences