Prism: Spectral Diversity for Multi-Agent Reinforcement Learning

Abstract

Parameter sharing is a key strategy in multi-agent reinforcement learning (MARL) for improving scalability, yet conventional fully shared architectures often collapse into homogeneous behaviors. Recent methods introduce diversity through clustering, pruning, or masking, but typically compromise resource efficiency. We propose Prism, a parameter sharing framework that induces spectral diversity by representing shared networks in the spectral domain via singular value decomposition (SVD). All agents share the singular vector directions while learning distinct spectral masks on singular values. This mechanism encourages inter-agent diversity and preserves scalability. Extensive experiments on both discrete (SMACv2) and continuous (MaMuJoCo) benchmarks show that Prism achieves competitive performance with superior resource efficiency.