TAIL: Task-specific Adapters for Imitation Learning with Large Pretrained Models
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Efficient Adaptation, Continual Learning, Robot Learning, Language-Conditioned Visuomotor Control, Few-shot Adaptation, Large Pretrained Models, Imitation Learning, Transfer Learning
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We propose a method of efficient adaptation for large pretrained decision-making models. Low-Rank Adaptation with our framework outperforms traditional fine-tuning with only 1% new parameters while avoiding issues like catastrophic forgetting.
Abstract: The full potential of large pretrained models remains largely untapped in control domains like robotics. This is mainly because of the scarcity of data and the computational challenges associated with training or fine-tuning these large models for such applications. Prior work mainly emphasizes either effective \emph{pretraining} of large models for decision-making or single-task adaptation. But real-world problems will require data-efficient, \emph{continual adaptation} for new control tasks. Recognizing these constraints, we introduce TAIL (Task-specific Adapters for Imitation Learning), a framework for efficient adaptation to new control tasks. Inspired by recent advancements in parameter-efficient fine-tuning in language domains, we explore efficient fine-tuning techniques---e.g., Bottleneck Adapters, P-Tuning, and Low-Rank Adaptation (LoRA)---in TAIL to adapt large pretrained models for new tasks with limited demonstration data. Our extensive experiments comparing prevalent parameter-efficient fine-tuning techniques and adaptation baselines suggest that TAIL with LoRA can achieve the best post-adaptation performance with only 1\% of the trainable parameters of full fine-tuning, while avoiding catastrophic forgetting and preserving adaptation plasticity in continual learning settings.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Primary Area: reinforcement learning
Submission Number: 4419
Loading