Multi-Class Unlearning for Image Classification via Weight Filtering

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Poppi, Samuele, Sarto, Sara, Cornia, Marcella, Baraldi, Lorenzo, Cucchiara, Rita
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929377515667456
author Poppi, Samuele
Sarto, Sara
Cornia, Marcella
Baraldi, Lorenzo
Cucchiara, Rita
author_facet Poppi, Samuele
Sarto, Sara
Cornia, Marcella
Baraldi, Lorenzo
Cucchiara, Rita
contents Machine Unlearning is an emerging paradigm for selectively removing the impact of training datapoints from a network. Unlike existing methods that target a limited subset or a single class, our framework unlearns all classes in a single round. We achieve this by modulating the network's components using memory matrices, enabling the network to demonstrate selective unlearning behavior for any class after training. By discovering weights that are specific to each class, our approach also recovers a representation of the classes which is explainable by design. We test the proposed framework on small- and medium-scale image classification datasets, with both convolution- and Transformer-based backbones, showcasing the potential for explainable solutions through unlearning.
format Preprint
id arxiv_https___arxiv_org_abs_2304_02049
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Multi-Class Unlearning for Image Classification via Weight Filtering
Poppi, Samuele
Sarto, Sara
Cornia, Marcella
Baraldi, Lorenzo
Cucchiara, Rita
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Machine Unlearning is an emerging paradigm for selectively removing the impact of training datapoints from a network. Unlike existing methods that target a limited subset or a single class, our framework unlearns all classes in a single round. We achieve this by modulating the network's components using memory matrices, enabling the network to demonstrate selective unlearning behavior for any class after training. By discovering weights that are specific to each class, our approach also recovers a representation of the classes which is explainable by design. We test the proposed framework on small- and medium-scale image classification datasets, with both convolution- and Transformer-based backbones, showcasing the potential for explainable solutions through unlearning.
title Multi-Class Unlearning for Image Classification via Weight Filtering
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2304.02049