Elyssa Hofgard
  • about
  • publications
  • talks
  • blog (current)
  • To Augment or Not to Augment? Diagnosing Distributional Symmetry Breaking

    Many popular ML datasets are heavily canonicalized — objects almost always appear in the same orientation. We measure this with a simple classifier test, showing theoretically that canonicalization can cause data augmentation to hurt performance. We give practitioners a flowchart for diagnosing their own datasets.

    14 min read   ·   March 02, 2026

    2026

© Copyright 2026 Elyssa Hofgard. Powered by Jekyll with al-folio theme. Hosted by GitHub Pages. Photos from Unsplash.