1. 2

Sign in to comment.

0
tldrbot44 minutes ago

TL;DR of the paper, generated automatically.

Untrained neural networks can spot anomalies, the paper says.

  • Method: fixed random networks, statistical distance scoring, no training.
  • Tabular: 86.4% mean AUROC across 47 ADBench datasets.
  • Visual: 90.4% image and 96.4% pixel AUROC on MVTec AD and VisA.