Summary of Splitz: Certifiable Robustness Via Split Lipschitz Randomized Smoothing, by Meiyu Zhong et al.
SPLITZ: Certifiable Robustness via Split Lipschitz Randomized Smoothing
by Meiyu Zhong, Ravi Tandon
First submitted to arxiv on: 3 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Information Theory (cs.IT)
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 approach, SPLITZ, combines two methods for certifiable robustness against adversarial examples: training classifiers with small Lipschitz constants and randomized smoothing. It splits a classifier into two halves, constrains the first half’s Lipschitz constant, and smooths the second half via randomization. This allows SPLITZ to exploit heterogeneity in Lipschitz constants across layers while inheriting the scalability of randomized smoothing. The approach is trained using a principled method and provides certified robustness guarantees during inference. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary SPLITZ is a new way to make sure that images are correctly classified even if they’re slightly changed. This is important because some images might be modified in ways that humans wouldn’t notice, but would change the computer’s prediction. There are two main approaches to do this: training models to have small “Lipschitz constants” and adding random noise to the image. SPLITZ combines these ideas by splitting a model into two parts, making sure one part is careful and the other part is more relaxed. This helps the model learn from its mistakes and make better predictions. |
Keywords
» Artificial intelligence » Inference
