Modelpedia
Work in progress
About
Findings
Models
Concepts
Methods
Datasets
Sources
Light
Dark
Sparse autoencoders reveal selective remapping of visual concepts during adaptation
2025-01-22
· ICLR 2025 Poster ·
anchor
Findings
IC-419
CLIP ViT-B/16's layer-11 residual stream contains class-discriminative information in sparse SAE latent directions, and ablating class-specific top-k latents significantly degrades zero-shot classification accuracy
IC-420
CLIP ViT-B/16's SAE latent interpretability is depth-dependent: layer 11 encodes semantic object concepts while layers 2, 5, and 8 encode local shapes and attention patterns