Summary of Improving Adversarial Training Using Vulnerability-aware Perturbation Budget, by Olukorede Fakorede et al.
Improving Adversarial Training using Vulnerability-Aware Perturbation Budget
by Olukorede Fakorede, Modeste Atsague, Jin Tian
First submitted to arxiv on: 6 Mar 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); 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 Adversarial Training (AT) is a technique used to improve the robustness of Deep Neural Networks (DNNs) to adversarial attacks. Typically, AT involves training DNN models with adversarial examples obtained within a pre-defined perturbation bound. However, crafting adversarial examples with fixed perturbation radius for all instances may not fully leverage the potency of AT. To address this limitation, we propose two simple and computationally efficient vulnerability-aware reweighting functions: Margin-Weighted Perturbation Budget (MWPB) and Standard-Deviation-Weighted Perturbation Budget (SDWPB). These methods assign perturbation radii to individual adversarial samples based on the vulnerability of their corresponding natural examples. Experimental results demonstrate that our proposed methods lead to genuine improvements in the robustness of AT algorithms against various adversarial attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making Deep Neural Networks (DNNs) more resistant to bad data. Bad data can trick DNNs into giving wrong answers, so we want to make them better at dealing with this kind of problem. One way to do this is called Adversarial Training (AT). Normally, when training a DNN for AT, we use fake examples that are slightly changed from the real ones. But what if these fake examples aren’t really representative of all the different types of bad data out there? We propose two new ways to create these fake examples that take into account how much each one is trying to trick the DNN. Our methods work better than usual AT and help make DNNs more robust against many types of bad data. |
