This document proposes using deep reinforcement learning to tune parameters in multi-agent simulations. Specifically, it uses DDPG (Deep Deterministic Policy Gradients), an actor-critic method for continuous control problems, to tune parameters. Additional components are needed, including an action converter to map actions to valid ranges, a redundant neural network actor to improve performance, and fixing the random seed to reduce variance between simulation runs. The proposed method is tested on an artificial market simulation and performs better than a Bayesian optimization baseline at tuning two parameters to match target skewness and kurtosis values. Future work could explore applying it to higher dimensional parameter tuning problems.
2022/10/30 BESC2022: Parameter Tuning Method for Multi-agent Simulation using Reinforcement Learning
1. Parameter Tuning Method for
Multi-agent Simulation using
Reinforcement Learning
Masanori HIRANO, Kiyoshi IZUMI
School of Engineering, The University of Tokyo
research@mhirano.jp
https://mhirano.jp/