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

While Federated learning (FL) is attractive for pulling privacy-preservingdistributed training data, the credibility of participating clients andnon-inspectable data pose new security threats, of which poisoning attacks areparticularly rampant and hard to defend without compromising privacy,performance or other desirable properties of FL. To tackle this problem, wepropose a self-purified FL (SPFL) method that enables benign clients to exploittrusted historical features of locally purified model to supervise the trainingof aggregated model in each iteration. The purification is performed by anattention-guided self-knowledge distillation where the teacher and studentmodels are optimized locally for task loss, distillation loss andattention-based loss simultaneously. SPFL imposes no restriction on thecommunication protocol and aggregator at the server. It can work in tandem withany existing secure aggregation algorithms and protocols for augmented securityand privacy guarantee. We experimentally demonstrate that SPFL outperformsstate-of-the-art FL defenses against various poisoning attacks. The attacksuccess rate of SPFL trained model is at most 3$\%$ above that of a cleanmodel, even if the poisoning attack is launched in every iteration with all butone malicious clients in the system. Meantime, it improves the model quality onnormal inputs compared to FedAvg, either under attack or in the absence of anattack.

SPFL: A Self-purified Federated Learning Method Against Poisoning Attacks