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

We study indiscriminate poisoning for linear learners where an adversaryinjects a few crafted examples into the training data with the goal of forcingthe induced model to incur higher test error. Inspired by the observation thatlinear learners on some datasets are able to resist the best known attacks evenwithout any defenses, we further investigate whether datasets can be inherentlyrobust to indiscriminate poisoning attacks for linear learners. For theoreticalGaussian distributions, we rigorously characterize the behavior of an optimalpoisoning attack, defined as the poisoning strategy that attains the maximumrisk of the induced model at a given poisoning budget. Our results prove thatlinear learners can indeed be robust to indiscriminate poisoning if theclass-wise data distributions are well-separated with low variance and the sizeof the constraint set containing all permissible poisoning points is alsosmall. These findings largely explain the drastic variation in empirical attackperformance of the state-of-the-art poisoning attacks on linear learners acrossbenchmark datasets, making an important initial step towards understanding theunderlying reasons some learning tasks are vulnerable to data poisoningattacks.

What Distributions are Robust to Indiscriminate Poisoning Attacks for Linear Learners?