The Price of Forgetting: Incentive Mechanism Design for Machine Unlearning

Data protection policies (e.g., GDPR) enforce the right to be forgotten and require companies to perform machine unlearning once users request data removal. This process incurs costs for server and degrades model performance, impacting users’ satisfaction. In this paper, we propose the first incentive mechanism for machine unlearning, where server compensates users to retain their data. We characterize server’s major unlearning costs, accuracy degradation and consumed time, in data redemption amount through experiments on three datasets and two unlearning algorithms. We model server-users interaction as a two-stage Stackelberg game. In Stage I, server optimizes compensation unit prices to minimize costs. In Stage II, users jointly decide data redemption amounts as a non-cooperative game. By restricting the feasible set of Stage I to Nash Equilibrium of Stage II, we formulate a challenging non-convex bilevel optimization problem. We propose an iterative algorithm to compute optimal unit prices in Stage I and equilibrium data redemption amounts in Stage II by characterizing bilevel problem’s convexity. We prove the distributed convergence of best response updates to the unique Nash equilibrium by showing Stage II is a submodular game. Experimental results show that our mechanism minimizes server cost and maximizes social welfare over two practical baselines.

• Publication year

  2025

• Venue

  IEEE Transactions on Mobile Computing

• Publication date

  2025-11-01

• Fields of study

  Computer Science

• Identifiers
  DOI 10.1109/TMC.2025.3582904
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• A Distributed Incentive Mechanism to Balance Demand and Communication Overhead for Multiple Federated Learning Tasks in IoV

  2025cited by this paper
• The Price of Forgetting: Data Redemption Mechanism Design for Machine Unlearning

  2024cited by this paper
• Incentive and Dynamic Client Selection for Federated Unlearning

  2024cited by this paper
• Communication Efficient and Provable Federated Unlearning

  2024cited by this paper
• Federated Unlearning: A Survey on Methods, Design Guidelines, and Evaluation Metrics

  2024cited by this paper
• Asynchronous Federated Unlearning

  2023cited by this paper
• PrivAim: A Dual-Privacy Preserving and Quality-Aware Incentive Mechanism for Federated Learning

  2023cited by this paper
• Cross-Entropy Loss Functions: Theoretical Analysis and Applications

  2023cited by this paper
• Incentive Mechanism Design for Federated Learning and Unlearning

  2023cited by this paper
• Strategic Data Revocation in Federated Unlearning

  2023cited by this paper
• The Impact of Adversarial Attacks on Federated Learning: A Survey

  2023cited by this paper
• Machine Unlearning: Solutions and Challenges

  2023cited by this paper
• PPA: Preference Profiling Attack Against Federated Learning

  2022cited by this paper
• A survey on federated learning: challenges and applications

  2022cited by this paper
• A Survey of Machine Unlearning

  2022cited by this paper
• Label‐only membership inference attacks on machine unlearning without dependence of posteriors

  2022cited by this paper
• Markov Chain Monte Carlo-Based Machine Unlearning: Unlearning What Needs to be Forgotten

  2022cited by this paper
• Incentive Mechanism Design for Joint Resource Allocation in Blockchain-Based Federated Learning

  2022cited by this paper
• Federated Learning With Fair Incentives and Robust Aggregation for UAV-Aided Crowdsensing

  2022cited by this paper
• Machine Unlearning of Features and Labels

  2021cited by this paper
• Adaptive Machine Unlearning

  2021cited by this paper
• A Novel Attribute Reconstruction Attack in Federated Learning

  2021cited by this paper
• Dynamic Games for Social Model Training Service Market via Federated Learning Approach

  2021cited by this paper
• When Machine Unlearning Jeopardizes Privacy

  2020influential reference
• Data and Model Dependencies of Membership Inference Attack

  2020cited by this paper
• Toward an Automated Auction Framework for Wireless Federated Learning Services Market

  2019cited by this paper
• Machine Unlearning

  2019influential reference
• Machine Learning from Theory to Algorithms: An Overview

  2018cited by this paper
• General Data Protection Regulation

  2018influential reference
• Mobile Data Trading: Behavioral Economics Analysis and Algorithm Design

  2017cited by this paper
• Dynamic Games Theory And Applications

  2016cited by this paper
• Spectrum Investment Under Uncertainty: A Behavioral Economics Perspective

  2016cited by this paper
• Mathematical Statistics : Basic Ideas and Selected Topics, Volumes I-II Package

  2015cited by this paper
• Utility-Optimal Random Access for Wireless Multimedia Networks

  2012cited by this paper
• Duopoly Competition in Dynamic Spectrum Leasing and Pricing

  2012cited by this paper
• Game theory.

  2011cited by this paper
• Convex optimization

  2010influential reference
• Learning Multiple Layers of Features from Tiny Images

  2009influential reference
• Analysis of Internet Users’ Level of Online Privacy Concerns

  2001influential reference
• Deterministic algorithms for constrained concave minimization: A unified critical survey

  1996cited by this paper
• A method for globally minimizing concave functions over convex sets

  1981cited by this paper
• Equilibrium Points in Nonzero-Sum n-Person Submodular Games

  1979cited by this paper
• The Levenberg-Marquardt algo-rithm: Implementation and theory

  1977cited by this paper
• You have printed the following article : Prospect Theory : An Analysis of Decision under Risk

  year unknowncited by this paper

• No citing papers are available for this paper.