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

With extensive studies on backdoor attack and detection, still fundamentalquestions are left unanswered regarding the limits in the adversary'scapability to attack and the defender's capability to detect. We believe thatanswers to these questions can be found through an in-depth understanding ofthe relations between the primary task that a benign model is supposed toaccomplish and the backdoor task that a backdoored model actually performs. Forthis purpose, we leverage similarity metrics in multi-task learning to formallydefine the backdoor distance (similarity) between the primary task and thebackdoor task, and analyze existing stealthy backdoor attacks, revealing thatmost of them fail to effectively reduce the backdoor distance and even forthose that do, still much room is left to further improve their stealthiness.So we further design a new method, called TSA attack, to automatically generatea backdoor model under a given distance constraint, and demonstrate that ournew attack indeed outperforms existing attacks, making a step closer tounderstanding the attacker's limits. Most importantly, we provide boththeoretic results and experimental evidence on various datasets for thepositive correlation between the backdoor distance and backdoor detectability,demonstrating that indeed our task similarity analysis help us betterunderstand backdoor risks and has the potential to identify more effectivemitigations.

Understanding Impacts of Task Similarity on Backdoor Attack and Detection