Preprint
Aug 2026
Feature Evolution and Migration during Vision Transformer Training
The experiments show that migration is concentrated early in training, occurs more often toward earlier layers than toward deeper layers, and declines as feature organization stabilizes, and the approach can be employed as a tool for understanding how ViTs learn and evolve.
Joonas Järve, Halil Ibrahim Aysel, Tarun Khajuria et al.
· 0 citations