Adaptive Machine Unlearning

Abstract

Data deletion algorithms aim to remove the influence of deleted data points from trained models at a cheaper computational cost than fully retraining those models. However, for sequences of deletions, most prior work in the non-convex setting gives valid guarantees only for sequences that are chosen independently of the models that are published. If people choose to delete their data as a function of the published models (because they don't like what the models reveal about them, for example), then the update sequence is adaptive. In this paper, we give a general reduction from deletion guarantees against adaptive sequences to deletion guarantees against non-adaptive sequences, using differential privacy and its connection to max information. Combined with ideas from prior work which give guarantees for non-adaptive deletion sequences, this leads to extremely flexible algorithms able to handle arbitrary model classes and training methodologies, giving strong provable deletion guarantees for adaptive deletion sequences. We show in theory how prior work for non-convex models fails against adaptive deletion sequences, and use this intuition to design a practical attack against the SISA algorithm of Bourtoule et al. [2021] on CIFAR-10, MNIST, Fashion-MNIST.

References in corpus (6)

• Algorithms that Remember: Model Inversion Attacks and Data Protection Law
• Generalization in Adaptive Data Analysis and Holdout Reuse
• Making AI Forget You: Data Deletion in Machine Learning
• Remember What You Want to Forget: Algorithms for Machine Unlearning
• Tempered Sigmoid Activations for Deep Learning with Differential Privacy
• Certified Data Removal from Machine Learning Models

Cited by in corpus (4)

• On the Effectiveness of Unlearning in Session-Based Recommendation
• SoK: Machine Learning Governance
• Feature Unlearning for Pre-trained GANs and VAEs
• An automatic differentiation system for the age of differential privacy