Summary of Discriminative Adversarial Unlearning, by Rohan Sharma et al.
Discriminative Adversarial Unlearning
by Rohan Sharma, Shijie Zhou, Kaiyi Ji, Changyou Chen
First submitted to arxiv on: 10 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 A novel machine unlearning framework is introduced, founded upon the min-max optimization paradigm. The approach leverages strong Membership Inference Attacks (MIA) to facilitate unlearning specific samples from a trained model. Two networks, an attacker and a defender, are pitted against each other in an adversarial objective, where the attacker aims to infer membership and the defender unlearns to defend while preserving performance. The algorithm can be trained end-to-end using backpropagation, following the iterative min-max approach. Additionally, a self-supervised objective is incorporated to address feature space discrepancies between the forget set and validation set, enhancing unlearning performance. The proposed method closely approximates retraining from scratch for both random sample forgetting and class-wise forgetting schemes on standard machine-unlearning datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to “forget” information in trained models is introduced. Imagine a game where one team tries to figure out which data points were used to train the model, while the other team tries to keep that information secret. The model can be updated repeatedly using an algorithm that combines two main parts: the attacker (trying to figure out the data) and the defender (trying to keep it secret). This approach is tested on various datasets and shows promising results in “forgetting” specific data points or classes, which could have important implications for applications like privacy protection. |
Keywords
* Artificial intelligence * Backpropagation * Inference * Optimization * Self supervised
