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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recently, the enactment of privacy regulations has promoted the rise of themachine unlearning paradigm. Existing studies of machine unlearning mainlyfocus on sample-wise unlearning, such that a learnt model will not exposeuser's privacy at the sample level. Yet we argue that such ability of selectiveremoval should also be presented at the attribute level, especially for theattributes irrelevant to the main task, e.g., whether a person recognized in aface recognition system wears glasses or the age range of that person. Througha comprehensive literature review, it is found that existing studies onattribute-related problems like fairness and de-biasing learning cannot addressthe above concerns properly. To bridge this gap, we propose a paradigm ofselectively removing input attributes from feature representations which wename `attribute unlearning'. In this paradigm, certain attributes will beaccurately captured and detached from the learned feature representations atthe stage of training, according to their mutual information. The particularattributes will be progressively eliminated along with the training proceduretowards convergence, while the rest of attributes related to the main task arepreserved for achieving competitive model performance. Considering thecomputational complexity during the training process, we not only give atheoretically approximate training method, but also propose an accelerationscheme to speed up the training process. We validate our method by spanningseveral datasets and models and demonstrate that our design can preserve modelfidelity and reach prevailing unlearning efficacy with high efficiency. Theproposed unlearning paradigm builds a foundation for future machine unlearningsystem and will become an essential component of the latest privacy-relatedlegislation.

Efficient Attribute Unlearning: Towards Selective Removal of Input Attributes from Feature Representations