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

Many backdoor removal techniques in machine learning models require cleanin-distribution data, which may not always be available due to proprietarydatasets. Model inversion techniques, often considered privacy threats, canreconstruct realistic training samples, potentially eliminating the need forin-distribution data. Prior attempts to combine backdoor removal and modelinversion yielded limited results. Our work is the first to provide a thoroughunderstanding of leveraging model inversion for effective backdoor removal byaddressing key questions about reconstructed samples' properties, perceptualsimilarity, and the potential presence of backdoor triggers.  We establish that relying solely on perceptual similarity is insufficient forrobust defenses, and the stability of model predictions in response to inputand parameter perturbations is also crucial. To tackle this, we introduce anovel bi-level optimization-based framework for model inversion, promotingstability and visual quality. Interestingly, we discover that reconstructedsamples from a pre-trained generator's latent space are backdoor-free, evenwhen utilizing signals from a backdoored model. We provide a theoreticalanalysis to support this finding. Our evaluation demonstrates that ourstabilized model inversion technique achieves state-of-the-art backdoor removalperformance without clean in-distribution data, matching or surpassingperformance using the same amount of clean samples.

Turning a Curse into a Blessing: Enabling In-Distribution-Data-Free Backdoor Removal via Stabilized Model Inversion