Research map / Multi-agent systems and games / Cooperative multi-agent RL

MA$^2$E: Addressing Partial Observability in Multi-Agent Reinforcement Learning with Masked Auto-Encoder

ICLR 2025 · Multi-agent systems and games › Cooperative multi-agent RL · approach: Decentralized actor-critic MARL

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes MA²E, a masked auto-encoder approach that helps cooperative multi-agent RL agents handle partial observability under the CTDE paradigm without direct communication.

Problem

Partial observability limits cooperation among agents in cooperative multi-agent reinforcement learning under centralized training with decentralized execution (CTDE); communication-based fixes can be restricted and add overhead.

Approach

Uses a masked auto-encoder to reconstruct missing or masked observation information so agents can infer unobserved aspects of the environment, instead of exchanging messages directly.

Experimental setup

Builds on: CTDE

Closest papers by meaning

• Enhancing Cooperative Multi-Agent Reinforcement Learning with State Modelling and Adversarial Exploration (ICML 2025 · similarity 0.90)
• Agent Modelling under Partial Observability for Deep Reinforcement Learning (NeurIPS 2021 · similarity 0.89)
• Deep Decentralized Multi-task Multi-Agent Reinforcement Learning under Partial Observability (ICML 2017 · similarity 0.89)
• M2I2: Learning Efficient Multi-Agent Communication via Masked State Modeling and Intention Inference (AAAI 2026 · similarity 0.89)
• Communication-efficient Multi-Agent Reinforcement Learning with Spatiotemporal Information Hub (AAAI 2026 · similarity 0.88)
• LLM-Guided Communication for Cooperative Multi-Agent Reinforcement Learning (ICML 2026 · similarity 0.88)