On Symmetry for Characterizing Agent Heterogeneity in Multi-Agent Systems
What it is
This paper defines a unified structural test for agent heterogeneity in finite-horizon Partially Observable Stochastic Games (POSGs) using policy-evaluation symmetries. It proves that these symmetries form a group where agent orbits define structural types, independent of specific policies or initial state distributions.
The problem it solves
Existing criteria for agent heterogeneity lack a unified structural test that is independent of specific policy choices or initial state distributions.
What you could use it for
- Designing shared versus agent-specific policy representations in multi-agent reinforcement learning
- Classifying structural agent types in cooperative Dec-POMDPs with common rewards
- Deriving exact value identities for weighted reward aggregation in multi-agent systems
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.