Summary of Persistent Classification: a New Approach to Stability Of Data and Adversarial Examples, by Brian Bell et al.
Persistent Classification: A New Approach to Stability of Data and Adversarial Examples
by Brian Bell, Michael Geyer, David Glickenstein, Keaton Hamm, Carlos Scheidegger, Amanda Fernandez, Juston Moore
First submitted to arxiv on: 11 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 The proposed framework for studying adversarial examples in classification problems focuses on understanding the differences between persistence metrics along interpolants of natural and adversarial points. The study shows that adversarial examples have significantly lower persistence than natural examples for large neural networks in the MNIST and ImageNet datasets. The researchers connect this lack of persistence with decision boundary geometry by measuring angles of interpolants with respect to decision boundaries, and also demonstrate the increase in robustness that can be achieved when training with a manifold alignment gradient metric. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Adversarial examples are tricky instances that can fool machine learning models into making wrong predictions. Researchers have been trying to understand why these examples exist and how they affect model performance. A new framework has been proposed to study adversarial examples, which looks at the stability of data points in a more detailed way than before. The results show that natural examples are much more persistent than adversarial ones when using large neural networks with popular datasets like MNIST and ImageNet. This lack of persistence is connected to how decision boundaries are formed in these models. The researchers also suggest a new way to train models that can make them more robust against adversarial attacks. |
Keywords
* Artificial intelligence * Alignment * Classification * Machine learning
