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

This work focuses on defending against the data poisoning based backdoorattacks, which bring in serious security threats to deep neural networks(DNNs). Specifically, given a untrustworthy training dataset, we aim to filterout potential poisoned samples, \ie, poisoned sample detection (PSD). The keysolution for this task is to find a discriminative metric between clean andpoisoned samples, even though there is no information about the potentialpoisoned samples (\eg, the attack method, the poisoning ratio). In this work,we develop an innovative detection approach from the perspective of thegradient \wrt activation (\ie, activation gradient direction, AGD) of eachsample in the backdoored model trained on the untrustworthy dataset. We presentan interesting observation that the circular distribution of AGDs among allsamples of the target class is much more dispersed than that of one cleanclass. Motivated by this observation, we firstly design a novel metric calledCosine similarity Variation towards Basis Transition (CVBT) to measure thecircular distribution's dispersion of each class. Then, we design a simple yeteffective algorithm with identifying the target class(es) using outlierdetection on CVBT scores of all classes, followed by progressively filtering ofpoisoned samples according to the cosine similarities of AGDs between everypotential sample and a few additional clean samples. Extensive experimentsunder various settings verify that given very few clean samples of each class,the proposed method could filter out most poisoned samples, while avoidingfiltering out clean samples, verifying its effectiveness on the PSD task. Codesare available athttps://github.com/SCLBD/bdzoo2/blob/dev/detection_pretrain/agpd.py.

Activation Gradient based Poisoned Sample Detection Against Backdoor Attacks