Sparse Autoencoders
Cross-source consensus on Sparse Autoencoders from 1 sources and 4 claims.
1 sources · 4 claims
Uses
How it works
Risks & contraindications
Comparisons
Highlighted claims
- Sparse autoencoders are used as a post-hoc method to decompose hidden activations into sparse interpretable features. — GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models
- TopK sparsity was chosen because BrainIAC representations satisfied most of the geometric criteria favoring it. — GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models
- Each SAE in the study reconstructed BrainIAC layer embeddings through sparse gated activations and unit-norm decoder columns. — GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models
- Standard TopK sparsity can cause feature mortality by giving zero gradient to features outside the selected top-k set. — GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models