Timezone: »

 
Adaptive Machine Unlearning
Varun Gupta · Christopher Jung · Seth Neel · Aaron Roth · Saeed Sharifi-Malvajerdi · Chris Waites

@
in Theory and Practice of Differential Privacy »

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.

Author Information

Varun Gupta (University of Pennsylvania)
Christopher Jung (University of Pennsylvania)
Seth Neel (Harvard)

PhD student at University of Pennsylvania

Aaron Roth (University of Pennsylvania)
Saeed Sharifi-Malvajerdi (University of Pennsylvania)
Chris Waites (Stanford University)

More from the Same Authors