Summary of Delta-influence: Unlearning Poisons Via Influence Functions, by Wenjie Li et al.
Delta-Influence: Unlearning Poisons via Influence Functions
by Wenjie Li, Jiawei Li, Christian Schroeder de Witt, Ameya Prabhu, Amartya Sanyal
First submitted to arxiv on: 20 Nov 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Cryptography and Security (cs.CR); Machine Learning (cs.LG)
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 -Influence approach utilizes influence functions to trace abnormal model behavior back to the responsible poisoned training data, even with only one affected example. This novel method applies data transformations that sever the link between poisoned training data and compromised test points without significantly affecting clean data. By detecting large negative shifts in influence scores following data transformations (influence collapse), -Influence accurately identifies poisoned training data, enabling effective unlearning through retraining. The authors validate their method across three vision-based poisoning attacks and three datasets, demonstrating consistent best performance among five unlearning strategies. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces a new way to identify the source of bad data in machine learning models. This is important because if we don’t know what’s causing problems, we can’t fix them. The authors created an algorithm called -Influence that uses existing techniques to find the bad data and then removes it from the model. They tested their method on three types of bad data attacks and showed that it works better than other approaches. |
Keywords
* Artificial intelligence * Machine learning
