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/

Zendy Plus

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

Zendy Tools

Zendy Open

Federated Learning (FL) is a promising technology that enables multipleactors to build a joint model without sharing their raw data. The distributednature makes FL vulnerable to various poisoning attacks, including modelpoisoning attacks and data poisoning attacks. Today, many byzantine-resilientFL methods have been introduced to mitigate the model poisoning attack, whilethe effectiveness when defending against data poisoning attacks still remainsunclear. In this paper, we focus on the most representative data poisoningattack - "label flipping attack" and monitor its effectiveness when attackingthe existing FL methods. The results show that the existing FL methods performsimilarly in Independent and identically distributed (IID) settings but fail tomaintain the model robustness in Non-IID settings. To mitigate the weaknessesof existing FL methods in Non-IID scenarios, we introduce the Honest ScoreClient Selection (HSCS) scheme and the corresponding HSCSFL framework. In theHSCSFL, The server collects a clean dataset for evaluation. Under eachiteration, the server collects the gradients from clients and then perform HSCSto select aggregation candidates. The server first evaluates the performance ofeach class of the global model and generates the corresponding risk vector toindicate which class could be potentially attacked. Similarly, the serverevaluates the client's model and records the performance of each class as theaccuracy vector. The dot product of each client's accuracy vector and globalrisk vector is generated as the client's host score; only the top p\% hostscore clients are included in the following aggregation. Finally, serveraggregates the gradients and uses the outcome to update the global model. Thecomprehensive experimental results show our HSCSFL effectively enhances the FLrobustness and defends against the "label flipping attack."

Honest Score Client Selection Scheme: Preventing Federated Learning Label Flipping Attacks in Non-IID Scenarios