VISR - Fast Task Inference with Variational Intrinsic Successor Features
This directory contains a Tensorflow-v1 / Sonnet implementation of the VISR algorithm in a notebook explaining how the approach can be used for task inference in a simple GridWorld. To launch the notebook in Google colab, click here.
VISR is a novel algorithm which learns controllable features that can be leveraged to provide enhanced generalization and fast task inference through the successor feature framework.
For details, see our paper Fast Task Inference with Variational Intrinsic Successor Features.
If you use the code here please cite this paper.
Steven Hansen, Will Dabney, Andre Barreto, David Warde-Farley, Tom Van de Wiele, Volodymyr Mnih. Fast Task Inference with Variational Intrinsic Successor Features. ICLR 2020. [arXiv].
Contributors
- Steven Hansen stevenhansen@google.com
- Will Dabney
- Andre Barreto
- David Warde-Farley
- Volodymyr Mnih
Disclaimer
This is not an official Google product.