Refactoring Policy for Compositional Generalizability using Self-Supervised Object Proposals
arXiv:2011.00971
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.
34th Conference on Neural Information Processing Systems (NeurIPS 2020), Vancouver, Canada
References in corpus (5)
- Fast Graph Representation Learning with PyTorch Geometric
- IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
- Multi-Object Representation Learning with Iterative Variational Inference
- MONet: Unsupervised Scene Decomposition and Representation
- Leveraging Procedural Generation to Benchmark Reinforcement Learning