NeurIPS 2020poster13 citations

Refactoring Policy for Compositional Generalizability using Self-Supervised Object Proposals

Tongzhou Mu, Jiayuan Gu, Zhiwei Jia, Hao Tang, Hao Su

Abstract

We study how to learn a policy with compositional generalizability. We propose a two-stage framework, which refactorizes a high-reward teacher policy into a generalizable student policy with strong inductive bias. Particularly, we implement an object-centric GNN-based student policy, whose input objects are learned from images through self-supervised learning. Empirically, we evaluate our approach on four difficult tasks that require compositional generalizability, and achieve superior performance compared to baselines.

BibTeX
@inproceedings{NEURIPS2020_64dcf3c5,
 author = {Mu, Tongzhou and Gu, Jiayuan and Jia, Zhiwei and Tang, Hao and Su, Hao},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {H. Larochelle and M. Ranzato and R. Hadsell and M.F. Balcan and H. Lin},
 pages = {8883--8894},
 publisher = {Curran Associates, Inc.},
 title = {Refactoring Policy for Compositional Generalizability using Self-Supervised Object Proposals},
 url = {https://proceedings.neurips.cc/paper_files/paper/2020/file/64dcf3c521a00dbb4d2a10a27a95a9d8-Paper.pdf},
 volume = {33},
 year = {2020}
}
Refactoring Policy for Compositional Generalizability using Self-Supervised Object Proposals · NeurIPS 2020