Machine unlearning
Techniques that let a model unlearn specific data without complete retraining, with no guarantee of total or verifiable forgetting.
Cao and Yang, "Towards Making Systems Forget with Machine Unlearning," IEEE Symposium on Security and Privacy, 2015: the first formulation of the problem, framed around the right to erasure in GDPR Article 17. Machine unlearning covers the techniques that let a learning model unlearn specific data by modifying its internal parameters, without the complete retraining that would be computationally prohibitive. Born about a decade ago, the discipline has produced so-called certified unlearning frameworks, which aim to guarantee mathematically, through probabilistic bounds, that the influence of the removed data has been sufficiently reduced. Its limits remain serious. No universally recognized method exists to verify that a model has actually forgotten, and an individual's data is entangled with that of thousands of others, so removing it without degradation is very delicate. Machine unlearning therefore offers a real but partial erasure, approximate and hard to prove, far from the clean deletion that the right to be forgotten assumes.
Textbook case: to answer an erasure request, a vendor applies a machine-unlearning technique to its model. It still cannot demonstrate to the regulator that all trace of the person has disappeared.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.