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 introduce techniques for exploring the functionality of a neural networkand extracting simple, human-readable approximations to its performance. Byperforming gradient ascent on the input space of the network, we are able toproduce large populations of artificial events which strongly excite a givenclassifier. By studying the populations of these events, we then directlyproduce what are essentially contour maps of the network's classificationfunction. Combined with a suite of tools for identifying the input dimensionsdeemed most important by the network, we can utilize these maps to efficientlyinterpret the dominant criteria by which the network makes its classification.  As a test case, we study networks trained to discriminate supersymmetric stopproduction in the dilepton channel from Standard Model backgrounds. In the caseof a heavy stop decaying to a light neutralino, we find individual neurons withlarge mutual information with $m_{T2}^{\ell\ell}$, a human-designed variablefor optimizing the analysis. The network selects events with significantmissing $p_T$ oriented azimuthally away from both leptons, efficientlyrejecting $t\overline{t}$ background. In the case of a light stop withthree-body decays to $Wb{\widetilde \chi}$ and little phase space, we findneurons that smoothly interpolate between a similar top-rejection strategy andan ISR-tagging strategy allowing for more missing momentum. We also find that aneural network trained on a stealth stop parameter point learns novel angularcorrelations.

Opening the black box of neural nets: case studies in stop/top discrimination