Summary of Adversarial Attacks on Hidden Tasks in Multi-task Learning, by Yu Zhe et al.
Adversarial Attacks on Hidden Tasks in Multi-Task Learning
by Yu Zhe, Rei Nagaike, Daiki Nishiyama, Kazuto Fukuchi, Jun Sakuma
First submitted to arxiv on: 24 May 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 This paper investigates the vulnerability of multi-task learning models to targeted attacks. Specifically, it explores the ability to compromise a model’s knowledge of a “hidden” task by leveraging information from other, related tasks and the shared network architecture. The proposed attack method, which exploits the model’s reliance on common representations across tasks, successfully degrades the accuracy of the target task while preserving performance on visible tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The researchers tested their method on two datasets: CelebA and DeepFashion. They found that their approach can effectively reduce the model’s ability to classify images in the hidden task, even when only non-target task information is available. This study contributes to our understanding of adversarial vulnerabilities in multi-task classifiers. |
Keywords
» Artificial intelligence » Multi task
