Certified Data Removal from Machine Learning Models
arXiv:1911.03030
Abstract
Good data stewardship requires removal of data at the request of the data's owner. This raises the question if and how a trained machine-learning model, which implicitly stores information about its training data, should be affected by such a removal request. Is it possible to "remove" data from a machine-learning model? We study this problem by defining certified removal: a very strong theoretical guarantee that a model from which data is removed cannot be distinguished from a model that never observed the data to begin with. We develop a certified-removal mechanism for linear classifiers and empirically study learning settings in which this mechanism is practical.
arXiv commentAccepted to ICML 2020
References in corpus (8)
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
- Deep Learning with Differential Privacy
- XLNet: Generalized Autoregressive Pretraining for Language Understanding
- YFCC100M: The New Data in Multimedia Research
- Understanding Black-box Predictions via Influence Functions
- Poisoning Attacks against Support Vector Machines
- Deep Learning with Dynamic Computation Graphs
- Making AI Forget You: Data Deletion in Machine Learning
Cited by in corpus (12)
- Descent-to-Delete: Gradient-Based Methods for Machine Unlearning
- Adaptive Machine Unlearning
- Variational Bayesian Unlearning
- Towards Probabilistic Verification of Machine Unlearning
- On the Necessity of Auditable Algorithmic Definitions for Machine Unlearning
- Certifiable Machine Unlearning for Linear Models
- SoK: Machine Learning Governance
- Bayesian Inference Forgetting
- Machine Unlearning via Algorithmic Stability
- Lightweight machine unlearning in neural network
- Selective Forgetting of Deep Networks at a Finer Level than Samples
- Near-Optimal Procedures for Model Discrimination with Non-Disclosure Properties