Policy Fine-tuning in Domain Adapation

What happens in neurons when we fine-tune a RL policy to adapt?

Introduction

This paper investigate the behavior of domain-adaption neurons in RL.

We also propose a method to make the training stable while training so that we can get better domain adaptation results with interpretable behaviors.

Related Work

Domain Adaptation

Neuron In RL

We leverage the two stage reinforcement learning where an agent has a domain-generalized policy and then, fine-tune the policy to a new environment. This work track the behavior of neurons while training.