Summary of Removing Spurious Correlation From Neural Network Interpretations, by Milad Fotouhi et al.
Removing Spurious Correlation from Neural Network Interpretations
by Milad Fotouhi, Mohammad Taha Bahadori, Oluwaseyi Feyisetan, Payman Arabshahi, David Heckerman
First submitted to arxiv on: 3 Dec 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Applications (stat.AP); Methodology (stat.ME)
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 proposed causal mediation approach in this paper addresses the issue of confounders in identifying neurons responsible for harmful behaviors. The current algorithms do not account for the impact of conversation topics, which can create spurious correlations. To overcome this limitation, the authors introduce a new method that controls for the effect of topic and demonstrates its effectiveness in experiments with two large language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us better understand how to identify the neurons behind harmful behaviors. Right now, our methods don’t consider the conversations we’re having, which can lead to mistakes. The researchers are fixing this by creating a new way to remove the impact of conversation topics and show it works well with two big language models. |
