Concept

Sparse Mapping

A method that maps the active features of one SAE into the feature space of another SAE through a correspondence relation, used to compare the consistency of attribution graphs across different training seeds.

Timeline (last 90 days)

Nov 3

The sparse mapping between features can explain 0.85 of a graph's energy on its active features, but when another seed's graph retains only its own features, this proportion drops to 0.47.

Unverified50%

All Facts (1)

Unverified

The sparse mapping between features can explain 0.85 of a graph's energy on its active features, but when another seed's graph retains only its own features, this proportion drops to 0.47.

50%

Source Articles

Sparse Autoencoder Attribution Graphs: Boundaries Decide EverythingNov 3