Summary of Shap Scores Fail Pervasively Even When Lipschitz Succeeds, by Olivier Letoffe et al.
SHAP scores fail pervasively even when Lipschitz succeeds
by Olivier Letoffe, Xuanxiang Huang, Joao Marques-Silva
First submitted to arxiv on: 18 Dec 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 recent widespread adoption of Shapley values in Explainable AI (XAI) has led to their common reference as SHAP scores. However, recent work highlighted examples of machine learning classifiers where computed SHAP scores can be thoroughly unsatisfactory, potentially misleading human decision-makers. This paper addresses these criticisms by exploring the issues with SHAP scores in Boolean and regression models. Specifically, it shows that there are arbitrarily many examples where SHAP scores must be deemed unsatisfactory for Boolean classifiers, and similar issues occur in regression models. Additionally, the paper investigates regression models respecting Lipschitz continuity, a measure crucial in model robustness, and demonstrates that SHAP score issues persist even with these models. Furthermore, it guarantees the existence of such issues for arbitrarily differentiable regression models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research looks at how well Shapley values work in explaining AI decisions. Right now, many people use Shapley values to understand why a machine learning model made a certain decision. However, some experts have shown that these values can be misleading and lead humans astray. This paper investigates this problem further by studying Boolean and regression models. The results show that there are many cases where the Shapley values don’t accurately explain the AI’s decision. This has important implications for how we use AI to make decisions. |
Keywords
» Artificial intelligence » Machine learning » Regression
