Feature Evolution and Migration during Vision Transformer Training

2026-08-20

AI Summary

研究通过稀疏自编码器提取特征并分析视觉变换器训练过程中的特征演变和迁移，揭示了特征在训练早期向浅层迁移并在深层更快稳定的现象。

Abstract

We present a novel view on feature evolution in Vision Transformers (ViTs) by visualizing the training process over two dimensions -- network depth (layer) and training time (epochs). We employ Sparse Autoencoders (SAEs) to extract candidate sparse features from CLS-token representations and compare their activation profiles across epoch--layer pairs. This allows us to study feature-level dynamics that are not directly visible from representation-level similarity measures. Furthermore, we demonstrate how this framework of feature evolution allows us to describe feature migration, the change in the layer where a feature is most detectable during training. Our 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. We further find that deeper layers stabilize earlier and more strongly than shallow layers. The results show that our approach can be employed as a tool for understanding how ViTs learn and evolve.

Problem

• Feature Evolution
• Vision Transformers
• Training Process
• Network Depth
• Feature Migration

Innovation

• Feature Evolution
• Vision Transformers
• Sparse Autoencoders
• Feature Migration