Summary of Leveraging Sub-optimal Data For Human-in-the-loop Reinforcement Learning, by Calarina Muslimani and Matthew E. Taylor
Leveraging Sub-Optimal Data for Human-in-the-Loop Reinforcement Learning
by Calarina Muslimani, Matthew E. Taylor
First submitted to arxiv on: 30 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Robotics (cs.RO)
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 Medium Difficulty summary: This paper proposes Sub-optimal Data Pre-training (SDP), an approach to improve the feedback efficiency of human-in-the-loop reinforcement learning (RL) methods. By pseudo-labeling low-quality data with rewards of zero, SDP pre-trains a reward model, enabling it to identify low-reward transitions without actual feedback. The authors demonstrate the effectiveness of SDP by comparing its performance with state-of-the-art HitL RL algorithms on nine robotic manipulation and locomotion tasks. SDP leverages scalar- and preference-based HitL RL algorithms and shows promising results in reducing the need for human interactions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This paper helps machines learn to do tasks better by giving them hints from humans. The problem is that it takes a lot of human effort to make these hints good enough. To fix this, the researchers created an approach called Sub-optimal Data Pre-training (SDP). SDP gives the machine a head start in learning what’s important and what’s not, so it doesn’t need as many human hints. The results show that SDP works well on nine different tasks, like robot manipulation and walking. |
Keywords
» Artificial intelligence » Reinforcement learning
