Descent-to-Delete: Gradient-Based Methods for Machine Unlearning

By Seth Neel, Aaron Roth, Saeed Sharifi-Malvajerdi. 2020

412 citations recorded by the source corpus.

We study the data deletion problem for convex models. By leveraging techniques from convex optimization and reservoir sampling, we give the first data deletion algorithms that are able to handle an arbitrarily long sequence of adversarial updates while promising both per-deletion run-time and steady-state error that do not grow with the length of the update sequence. We also introduce several new conceptual distinctions: for example, we can ask that after a deletion, the entire state maintained by the optimization algorithm is statistically indistinguishable from the state that would have resulted had we retrained, or we can ask for the weaker condition that only the observable output is statistically indistinguishable from the observable output that would have resulted from retraining. We are able to give more efficient deletion algorithms under this weaker deletion criterion.

Subfields: Artificial Intelligence

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

Citations in the corpus

Learning with Recoverable Forgetting (2022, positive): “[26] give the first data deletion algorithms.”
Mixed-Privacy Forgetting in Deep Networks (2020, methodology): “Recently, [35] proposed a gradient descent based method for data deletion in convex settings, with theoretical guarantees for multiple forgetting requests.”
Measuring Forgetting of Memorized Training Examples (2022, neutral): “Algorithms for machine unlearning have been proposed for k-means [24], empirical risk minimization [29, 35, 59, 76, 67], and deep learning [18, 26, 25, 60, 5].”
SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation (2023, background): “MU aims at modifying ML models to eliminate the influence of specific data points or classes, initially developed to mitigate potential privacy breaches post-training (Ginart et al., 2019; Neel et al., 2021; Ullah et al., 2021; Sekhari et al., 2021).”
A Survey of Machine Unlearning (2022, neutral): “have some privacy leak, even though some provable guarantees are provided for special cases [57, 97, 106].”
Towards Unbounded Machine Unlearning (2023, background): “Sekhari et al. [2021] and Neel et al. [2021] adopt similar DP-inspired deﬁnitions of (approximate) unlearning based on the goal of indistinguishability from retrain-from-scratch.”
Rethinking machine unlearning for large language models (2024, neutral): “…performance, the landscape of MU has expanded to encompass diverse domains, such as image classification Ginart et al. (2019); Golatkar et al. (2020) In the literature, ‘exact’ unlearning, which involves retraining the model from scratch after removing specific training data points, is often…”
Adaptive Machine Unlearning (2021, methodology): “These update requests are formally defined below, similar to how they are defined in [Neel et al., 2021].”
Machine Unlearning of Features and Labels (2021, methodology): “[50] perform a similar analysis for gradient-based update strategies.”
Large Language Model Unlearning (2023, neutral): “…for classification models, including data-reversed training [Tarun et al., 2023, Liu et al., 2022, Chundawat et al., 2023], optimization-based unlearning [Guo et al., 2019, Neel et al., 2021] and influence function based approaches [Jia et al., 2023, Warnecke et al., 2021, Izzo et al., 2021].”
Remember What You Want to Forget: Algorithms for Machine Unlearning (2021, methodology): “This relaxation has inspired the design of several eﬃcient algorithms for data deletion from machine learning models (Guo et al., 2020; Izzo et al., 2021; Neel et al., 2021; Ullah et al., 2021).”