Summary of ‘explaining Rl Decisions with Trajectories’: a Reproducibility Study, by Karim Abdel Sadek et al.
‘Explaining RL Decisions with Trajectories’: A Reproducibility Study
by Karim Abdel Sadek, Matteo Nulli, Joan Velja, Jort Vincenti
First submitted to arxiv on: 11 Nov 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
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 paper investigates the reproducibility of a novel approach to explainable reinforcement learning introduced in the original paper ‘Explaining RL decisions with trajectories’. The approach attributes decision-making processes to specific clusters of trajectories encountered during training. The main claims from the paper are verified, including the effects of training on less trajectories and the influence of distant trajectories on agent decisions. Additionally, quantitative metrics are introduced to further support these claims, and experiments are extended to new environments. While some claims hold up, others require further investigation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how well a special way of understanding why a computer makes certain decisions works when someone tries to copy it. This approach helps explain what the computer is doing by looking at groups of steps it took to get to its decision. The original paper said some things about this method, and this new work checks if those things are true. It also looks at how well different ways of grouping these steps work and even introduces a special way of measuring how good this approach is. While some of the claims from the original paper hold up, others need more study. |
Keywords
» Artificial intelligence » Reinforcement learning
