Randomly masking and predicting word tokens has been a successful approach in pre-training language models for a variety of downstream tasks. In this work, we observe that the same idea also applies naturally to sequential decision-making, where many well-studied tasks like behavior cloning, offline reinforcement learning, inverse dynamics, and waypoint conditioning correspond to different sequence maskings over a sequence of states, actions, and returns. We introduce the UniMASK framework, which provides a unified way to specify models which can be trained on many different sequential decision-making tasks. We show that a single UniMASK model is often capable of carrying out many tasks with performance similar to or better than single-task models. Additionally, after fine-tuning, our UniMASK models consistently outperform comparable single-task models. Our code is publicly available at https://github.com/micahcarroll/uniMASK.
@article{arxiv.2211.10869,
title = {UniMASK: Unified Inference in Sequential Decision Problems},
author = {Micah Carroll and Orr Paradise and Jessy Lin and Raluca Georgescu and Mingfei Sun and David Bignell and Stephanie Milani and Katja Hofmann and Matthew Hausknecht and Anca Dragan and Sam Devlin},
journal= {arXiv preprint arXiv:2211.10869},
year = {2022}
}
Comments
NeurIPS 2022 (Oral). A prior version was published at an ICML Workshop, available at arXiv:2204.13326