Summary of Reinforcement Learning Gradients As Vitamin For Online Finetuning Decision Transformers, by Kai Yan et al.
Reinforcement Learning Gradients as Vitamin for Online Finetuning Decision Transformers
by Kai Yan, Alexander G. Schwing, Yu-Xiong Wang
First submitted to arxiv on: 31 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The abstract discusses recent advancements in Decision Transformers for offline Reinforcement Learning (RL), which complete trajectories autoregressively. While improvements have been made, the online finetuning of these models has received surprisingly little attention. The state-of-the-art Online Decision Transformer (ODT) still struggles when pre-trained with low-reward offline data. This paper analyzes the theoretical limitations of online-finetuning and shows that the Return-To-Go (RTG) metric used to evaluate performance can hinder fine-tuning. Instead, incorporating TD3 gradients into the ODT finetuning process can significantly improve performance, especially when using pre-trained models with low-reward offline data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Decision Transformers have revolutionized offline Reinforcement Learning by completing trajectories in an autoregressive way. However, online finetuning of these models has been surprisingly under-explored. The state-of-the-art Online Decision Transformer (ODT) still struggles when pre-trained with low-reward offline data. This paper shows that the RTG metric used to evaluate performance can hinder fine-tuning and suggests a simple solution: adding TD3 gradients to the finetuning process of ODT. |
Keywords
» Artificial intelligence » Attention » Autoregressive » Fine tuning » Reinforcement learning » Transformer
