Few-Shot Unlearning by Model Inversion
This addresses a practical challenge in machine unlearning for scenarios where training data is inaccessible, offering a solution for erasing data to correct errors or protect privacy.
The paper tackles the problem of machine unlearning when only a few samples of the target data are available, proposing a framework that uses model inversion to retrieve a proxy of training data and adjust it based on unlearning intentions, achieving results that outperform state-of-the-art methods even with full target data.
We consider a practical scenario of machine unlearning to erase a target dataset, which causes unexpected behavior from the trained model. The target dataset is often assumed to be fully identifiable in a standard unlearning scenario. Such a flawless identification, however, is almost impossible if the training dataset is inaccessible at the time of unlearning. Unlike previous approaches requiring a complete set of targets, we consider few-shot unlearning scenario when only a few samples of target data are available. To this end, we formulate the few-shot unlearning problem specifying intentions behind the unlearning request (e.g., purely unlearning, mislabel correction, privacy protection), and we devise a straightforward framework that (i) retrieves a proxy of the training data via model inversion fully exploiting information available in the context of unlearning; (ii) adjusts the proxy according to the unlearning intention; and (iii) updates the model with the adjusted proxy. We demonstrate that our method using only a subset of target data can outperform the state-of-the-art unlearning methods even with a complete indication of target data.