Summary of Gaussian Smoothing in Saliency Maps: the Stability-fidelity Trade-off in Neural Network Interpretability, by Zhuorui Ye et al.
Gaussian Smoothing in Saliency Maps: The Stability-Fidelity Trade-Off in Neural Network Interpretability
by Zhuorui Ye, Farzan Farnia
First submitted to arxiv on: 6 Nov 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 explores the stability and faithfulness of saliency maps, specifically gradient-based methods like Smooth-Grad, Simple-Grad, and Integrated-Gradients. The authors investigate how Gaussian smoothing affects these maps’ ability to interpret neural network classifiers and identify phenomena from their learned functions. They prove theoretical bounds on the stability error of each method and analyze the impact of Gaussian smoothing on map fidelity. Empirical experiments on standard image datasets confirm the predicted trade-off between stability and faithfulness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how computers learn to recognize things. It looks at special maps that show which parts of an image helped a computer decide what it saw. But these maps can be tricky because they depend on the way the computer was trained, which can be random sometimes. The researchers wanted to know if adding some extra noise (Gaussian smoothing) would make the maps more stable and reliable. They proved mathematically that this works, but also found that too much noise makes the maps less accurate. This is important for making sure computers understand things correctly. |
Keywords
» Artificial intelligence » Neural network
