Summary of Manifold Integrated Gradients: Riemannian Geometry For Feature Attribution, by Eslam Zaher et al.
Manifold Integrated Gradients: Riemannian Geometry for Feature Attribution
by Eslam Zaher, Maciej Trzaskowski, Quan Nguyen, Fred Roosta
First submitted to arxiv on: 16 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Human-Computer Interaction (cs.HC); Differential Geometry (math.DG)
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 investigates the reliability concerns of Integrated Gradients (IG), a widely used feature attribution method for deep learning models. IG is prone to generating noisy visualizations and vulnerable to adversarial attacks. To address these issues, the authors propose an adaptation of path-based feature attribution that aligns with the intrinsic geometry of the data manifold. Experiments on real-world image datasets using deep generative models demonstrate the effectiveness of this approach in producing more intuitive explanations and increasing robustness to targeted attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure a special tool called Integrated Gradients (IG) works correctly. IG helps us understand what parts of an image are important for a computer to recognize it. But sometimes, IG gets confused or gives wrong answers. The researchers found two main problems with IG: it makes messy pictures and can be tricked into giving bad answers. They came up with a new way to use IG that matches the natural patterns in the data, making it better at explaining things and harder for bad actors to fool. |
Keywords
» Artificial intelligence » Deep learning
