Summary of Towards Class-wise Robustness Analysis, by Tejaswini Medi et al.
Towards Class-wise Robustness Analysis
by Tejaswini Medi, Julia Grabinski, Margret Keuper
First submitted to arxiv on: 29 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
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 limitations of deep neural networks in real-life scenarios due to their susceptibility to domain shifts and adversarial attacks. Despite their success in solving many downstream tasks, these models are vulnerable to common corruptions and adversarial examples, significantly reducing their performance. The authors focus on class-wise differences in robustness, which are critical for developing robust neural architectures. They analyze the latent space structures of adversarially trained robust classification models and assess their strong and weak properties across different classes. The study finds that the number of false positives of target classes impacts their vulnerability to attacks, making it essential to evaluate each class’s susceptibility to misclassification. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Deep learning can be very good at recognizing images, but there are limitations. When the environment changes or someone tries to trick the system, deep neural networks can fail. This is a problem because we want these systems to work well in real-life situations. The authors of this paper looked at how different classes (like cats and dogs) react when they’re attacked with fake data or corrupted images. They found that some classes are more vulnerable than others, which means it’s important to evaluate each class separately to understand its weaknesses. |
Keywords
» Artificial intelligence » Classification » Deep learning » Latent space
