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

Due to the distributed nature of Federated Learning (FL), researchers haveuncovered that FL is vulnerable to backdoor attacks, which aim at injecting asub-task into the FL without corrupting the performance of the main task.Single-shot backdoor attack achieves high accuracy on both the main task andbackdoor sub-task when injected at the FL model convergence. However, theearly-injected single-shot backdoor attack is ineffective because: (1) themaximum backdoor effectiveness is not reached at injection because of thedilution effect from normal local updates; (2) the backdoor effect decreasesquickly as the backdoor will be overwritten by the newcoming normal localupdates. In this paper, we strengthen the early-injected single-shot backdoorattack utilizing FL model information leakage. We show that the FL convergencecan be expedited if the client trains on a dataset that mimics the distributionand gradients of the whole population. Based on this observation, we proposed atwo-phase backdoor attack, which includes a preliminary phase for thesubsequent backdoor attack. In the preliminary phase, the attacker-controlledclient first launches a whole population distribution inference attack and thentrains on a locally crafted dataset that is aligned with both the gradient andinferred distribution. Benefiting from the preliminary phase, the laterinjected backdoor achieves better effectiveness as the backdoor effect will beless likely to be diluted by the normal model updates. Extensive experimentsare conducted on MNIST dataset under various data heterogeneity settings toevaluate the effectiveness of the proposed backdoor attack. Results show thatthe proposed backdoor outperforms existing backdoor attacks in both successrate and longevity, even when defense mechanisms are in place.

Technical Report: Assisting Backdoor Federated Learning with Whole Population Knowledge Alignment