Summary of Do Concept Bottleneck Models Respect Localities?, by Naveen Raman et al.
Do Concept Bottleneck Models Respect Localities?
by Naveen Raman, Mateo Espinosa Zarlenga, Juyeon Heo, Mateja Jamnik
First submitted to arxiv on: 2 Jan 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 paper investigates the faithfulness of Concept Bottleneck Models (CBMs) in capturing localities in datasets. CBMs are popular architectures that explain model predictions using human-understandable concepts. However, it is unclear whether these models accurately predict underlying concepts. The authors examine how CBM predictions change when perturbing model inputs and find that CBMs may not capture localities even when independent concepts are localized to non-overlapping feature subsets. This can lead to accurate but uninterpretable models that fail to learn localities. The study highlights the fragility of CBM interpretability, as CBMs occasionally rely on spurious features. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well Concept Bottleneck Models (CBMs) understand and explain what they’re predicting. CBMs are special types of AI models that help people understand why they made certain predictions. But it’s not clear if these models really get what’s going on underneath the surface. The researchers looked at how well CBMs work when you change the input data a little bit, and found that sometimes they don’t really capture what’s important. This can make their predictions correct but hard to understand. |
