Skip to content

[Proposal] Support data whitening in SAE training #647

@chanind

Description

@chanind

Data Whitening Improves Sparse Autoencoder Learning shows that whitening the input data during training results in better SAEs. This seems like a pretty intuitive and sort of obvious thing to do, it's surprising nobody has done this so far. We already support normalizing activations using normalize_activations="expected_average_only_in", we could add an additional option "covariance_whitening" implementing the technique from this paper.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions