Summary of Optimizing Latent Goal by Learning From Trajectory Preference, By Guangyu Zhao et al.
Optimizing Latent Goal by Learning from Trajectory Preference
by Guangyu Zhao, Kewei Lian, Haowei Lin, Haobo Fu, Qiang Fu, Shaofei Cai, Zihao Wang, Yitao Liang
First submitted to arxiv on: 3 Dec 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Machine Learning (cs.LG)
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 This research proposes a framework called Preference Goal Tuning (PGT) to improve the performance of instruction-following policies for open-world agents. The existing approaches are highly susceptible to the initial prompt, requiring extra efforts in selecting the best instructions. PGT allows the policy to interact with the environment, collecting trajectories that are categorized into positive and negative samples based on preference. The framework then fine-tunes the initial goal latent representation using preference learning while keeping the policy backbone frozen. The experiments show that PGT achieves significant improvements over existing approaches, achieving an average relative improvement of 72.0% and 81.6% across 17 tasks in two foundation policies. Additionally, PGT surpasses full fine-tuning in out-of-distribution task-execution environments by 13.4%, demonstrating robustness to environment changes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research helps machines understand human intentions better. They use a new approach called Preference Goal Tuning (PGT) that lets machines learn from their experiences and make good decisions. Before, machines had trouble understanding what humans wanted them to do because they relied too much on the first instruction they received. PGT changes this by letting machines collect more information and adjust their goals based on what works best. The results show that PGT makes machines smarter and more capable of completing tasks. |
Keywords
» Artificial intelligence » Fine tuning » Prompt
