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/

Zendy Plus

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

Zendy Tools

Zendy Open

Researchers have developed neural network verification algorithms motivatedby the need to characterize the robustness of deep neural networks. Theverifiers aspire to answer whether a neural network guarantees certainproperties with respect to all inputs in a space. However, many verifiersinaccurately model floating point arithmetic but do not thoroughly discuss theconsequences.  We show that the negligence of floating point error leads to unsoundverification that can be systematically exploited in practice. For a pretrainedneural network, we present a method that efficiently searches inputs aswitnesses for the incorrectness of robustness claims made by a completeverifier. We also present a method to construct neural network architecturesand weights that induce wrong results of an incomplete verifier. Our resultshighlight that, to achieve practically reliable verification of neuralnetworks, any verification system must accurately (or conservatively) model theeffects of any floating point computations in the network inference orverification system.

Exploiting Verified Neural Networks via Floating Point Numerical Error