- Published on
GAUGE-INVARIANT REPRESENTATION HOLONOMY
- Authors
- Name
- Vishal V
- @VishalVignesh_
Notes
ABSTRACT
Existing similarity measures such as CKA or SVCCA capture pointwise overlap between activation sets, but miss how representations change along input paths ()
We introduce representation holonomy, a gauge-invariant statistic that measures this path dependence ()
Empirically, holonomy increases with loop radius, separates models that appear similar under CKA, and correlates with adversarial and corruption robustness ()