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

Few-shot classifiers have been shown to exhibit promising results in usecases where user-provided labels are scarce. These models are able to learn topredict novel classes simply by training on a non-overlapping set of classes.This can be largely attributed to the differences in their mechanisms ascompared to conventional deep networks. However, this also offers newopportunities for novel attackers to induce integrity attacks against suchmodels, which are not present in other machine learning setups. In this work,we aim to close this gap by studying a conceptually simple approach to defendfew-shot classifiers against adversarial attacks. More specifically, we proposea simple attack-agnostic detection method, using the concept of self-similarityand filtering, to flag out adversarial support sets which destroy theunderstanding of a victim classifier for a certain class. Our extendedevaluation on the miniImagenet (MI) and CUB datasets exhibit good attackdetection performance, across three different few-shot classifiers and acrossdifferent attack strengths, beating baselines. Our observed results allow ourapproach to establishing itself as a strong detection method for support setpoisoning attacks. We also show that our approach constitutes a generalizableconcept, as it can be paired with other filtering functions. Finally, weprovide an analysis of our results when we vary two components found in ourdetection approach.

Towards A Conceptually Simple Defensive Approach for Few-shot classifiers Against Adversarial Support Samples