Summary of Beyond Concept Bottleneck Models: How to Make Black Boxes Intervenable?, by Sonia Laguna et al.
Beyond Concept Bottleneck Models: How to Make Black Boxes Intervenable?
by Sonia Laguna, Ričards Marcinkevičs, Moritz Vandenhirtz, Julia E. Vogt
First submitted to arxiv on: 24 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 This paper introduces a method to perform concept-based interventions on pretrained neural networks, which are not interpretable by design. The authors formalize the notion of intervenability as a measure of the effectiveness of concept-based interventions and use this definition to fine-tune black boxes. They explore the intervenability of black-box classifiers on synthetic tabular and natural image benchmarks using backbone architectures of varying complexity. The proposed fine-tuning improves intervention effectiveness and often yields better-calibrated predictions. The authors apply their techniques to deep chest X-ray classifiers, showing that fine-tuned black boxes are more intervenable than concept bottleneck models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how we can make AI models like neural networks easier to work with by giving them instructions or “interventions” based on the concepts they use. The authors created a way to do this using only a small amount of labeled data, which is important because many AI models are not designed to be understood. They tested their method on different types of data and showed that it works well even when the labels come from other sources like language models. This could help us use AI models in new ways, like making medical diagnoses or analyzing X-rays. |
Keywords
* Artificial intelligence * Fine tuning
