English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

The increased attention to regulating the outputs of deep generative models,driven by growing concerns about privacy and regulatory compliance, hashighlighted the need for effective control over these models. This necessityarises from instances where generative models produce outputs containingundesirable, offensive, or potentially harmful content. To tackle thischallenge, the concept of machine unlearning has emerged, aiming to forgetspecific learned information or to erase the influence of undesired datasubsets from a trained model. The objective of this work is to prevent thegeneration of outputs containing undesired features from a pre-trained GANwhere the underlying training data set is inaccessible. Our approach isinspired by a crucial observation: the parameter space of GANs exhibitsmeaningful directions that can be leveraged to suppress specific undesiredfeatures. However, such directions usually result in the degradation of thequality of generated samples. Our proposed method, known as'Adapt-then-Unlearn,' excels at unlearning such undesirable features while alsomaintaining the quality of generated samples. This method unfolds in twostages: in the initial stage, we adapt the pre-trained GAN using negativesamples provided by the user, while in the subsequent stage, we focus onunlearning the undesired feature. During the latter phase, we train thepre-trained GAN using positive samples, incorporating a repulsion regularizer.This regularizer encourages the model's parameters to be away from theparameters associated with the adapted model from the first stage while alsomaintaining the quality of generated samples. To the best of our knowledge, ourapproach stands as first method addressing unlearning in GANs. We validate theeffectiveness of our method through comprehensive experiments.

Adapt then Unlearn: Exploiting Parameter Space Semantics for Unlearning in Generative Adversarial Networks