Summary of Decompose the Model: Mechanistic Interpretability in Image Models with Generalized Integrated Gradients (gig), by Yearim Kim et al.
Decompose the model: Mechanistic interpretability in image models with Generalized Integrated Gradients (GIG)
by Yearim Kim, Sangyu Han, Sangbum Han, Nojun Kwak
First submitted to arxiv on: 3 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- 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 This paper proposes a novel approach to explainable AI (XAI) in image models, bridging the gap between local explanations of individual decisions and global explanations with high-level concepts. The authors introduce Pointwise Feature Vectors (PFVs) and Effective Receptive Fields (ERFs) to decompose model embeddings into interpretable Concept Vectors, which are then analyzed using Generalized Integrated Gradients (GIG). This allows for a comprehensive, dataset-wide understanding of image models’ operational mechanics. The approach is validated through both qualitative and quantitative evaluations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how image recognition models work by showing us what they’re looking at. Right now, we can only see what one part of the model is doing, not the whole thing. This new method lets us see the whole process from start to finish, which is important because it helps us figure out why the model is making certain decisions. The authors use a special kind of math called concept vectors to break down the model’s thinking into smaller parts we can understand. |
