Interpretable Discovery in Large Image Data Sets

  • 2018-06-21 17:30:26
  • Kiri L. Wagstaff, Jake Lee
  • 9

Abstract

Automated detection of new, interesting, unusual, or anomalous images withinlarge data sets has great value for applications from surveillance (e.g.,airport security) to science (observations that don't fit a given theory canlead to new discoveries). Many image data analysis systems are turning toconvolutional neural networks (CNNs) to represent image content due to theirsuccess in achieving high classification accuracy rates. However, CNNrepresentations are notoriously difficult for humans to interpret. We describea new strategy that combines novelty detection with CNN image features toachieve rapid discovery with interpretable explanations of novel image content.We applied this technique to familiar images from ImageNet as well as to ascientific image collection from planetary science.

 

Quick Read (beta)

loading the full paper ...