mike-gimelfarb / contextual-policy-reuse-deep-rl

Framework for Contextually Transferring Knowledge from Multiple Source Policies in Deep Reinforcement Learning

Contextual Policy Reuse using Deep Mixture Models

This repository will contain the code to reproduce experiments in an upcoming paper with the same name. The main idea of this framework is to enable contextual reuse of multiple source policies in deep reinforcement learning problems with similar goals but different dynamics.

Code will be uploaded shortly: stay tuned.

Citation

To cite the framework:

@misc{gimelfarb2020contextual,
title={Contextual Policy Reuse using Deep Mixture Models},
author={Michael Gimelfarb and Scott Sanner and Chi-Guhn Lee},
year={2020},
eprint={2003.00203},
archivePrefix={arXiv},
primaryClass={cs.LG} }