Dynamic Noises of Multi-Agent Environments Can Improve Generalization: Agent-based Models meets Reinforcement Learning

TL;DR

This paper demonstrates that the inherent stochasticity in agent-based models used as environments for reinforcement learning can enhance the generalization capabilities of RL agents across diverse scenarios.

Contribution

It introduces the idea that the non-deterministic dynamics of ABMs can be beneficial for RL generalization, supported by empirical evidence from epidemic control simulations.

Findings

ABM-based environments improve RL reward performance.

Non-deterministic dynamics enhance generalization across parameters.

ABMs offer microfoundational insights despite higher computational costs.

Abstract

We study the benefits of reinforcement learning (RL) environments based on agent-based models (ABM). While ABMs are known to offer microfoundational simulations at the cost of computational complexity, we empirically show in this work that their non-deterministic dynamics can improve the generalization of RL agents. To this end, we examine the control of an epidemic SIR environments based on either differential equations or ABMs. Numerical simulations demonstrate that the intrinsic noise in the ABM-based dynamics of the SIR model not only improve the average reward but also allow the RL agent to generalize on a wider ranges of epidemic parameters.