Unrolling SGD: Understanding Factors Influencing Machine Unlearning

By Anvith Thudi, Gabriel Deza, Varun Chandrasekaran, Nicolas Papernot. European Symposium on Security and Privacy, 2021

342 citations recorded by the source corpus.

Machine unlearning is the process through which a deployed machine learning model is made to forget about some of its training data points. While naively retraining the model from scratch is an option, it is almost always associated with large computational overheads for deep learning models. Thus, several approaches to approximately unlearn have been proposed along with corresponding metrics that formalize what it means for a model to forget about a data point. In this work, we first taxonomize approaches and metrics of approximate unlearning. As a result, we identify verification error, i.e., the $\ell_{2}$ difference between the weights of an approximately unlearned and a naively retrained model, as an approximate unlearning metric that should be optimized for as it subsumes a large class of other metrics. We theoretically analyze the canonical training algorithm, stochastic gradient descent (SGD), to surface the variables which are relevant to reducing the verification error of approximate unlearning for SGD. From this analysis, we first derive an easy-to-compute proxy for verification error (termed unlearning error). The analysis also informs the design of a new training objective penalty that limits the overall change in weights during SGD and as a result facilitates approximate unlearning with lower verification error. We validate our theoretical work through an empirical evaluation on learning with CIFAR-10, CIFAR-100, and IMDB sentiment analysis.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/f3a8de8c9117926199b0da0b548fecae3e20793a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Yihua Zhang
2. Jiancheng Liu
3. Eric Wong
4. Sijia Liu
5. Yuanshun Yao
6. Song Mei
7. Yu Bai
8. Licong Lin

In Computer Vision and Pattern Recognition

1. Yihua Zhang
2. Jiancheng Liu
3. Eric Wong
4. Sijia Liu
5. Yuanshun Yao
6. Song Mei
7. Yu Bai
8. Licong Lin

Citations in the corpus

• Large Language Model Unlearning (2023, background): “For example, a typical optimization-based techinque [Thudi et al., 2022] is gradient ascent (GA).”
• Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning (2024, background): “Since its proposal by Cao & Yang (2015), machine unlearning has been extensively studied in the classification literature (Bourtoule et al., 2021; Golatkar et al., 2020; Ginart et al., 2019; Thudi et al., 2022; Izzo et al., 2021; Koh & Liang, 2017; Guo et al., 2019; Sekhari et al., 2021).”
• SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation (2023, methodology): “4, FT (Warnecke et al., 2021), RL (Golatkar et al., 2020), GA (Thudi et al., 2022a), IU (Izzo et al., 2021; Jia et al., 2023), ℓ 1 -sparse (Jia et al., 2023), as well as 4 new baselines , including 2 boundary unlearning methods (Chen et al., 2023), boundary shrink ( BS ) and boundary expanding ( BE…”