4 papers
Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds
Thomas Fel, Matthew Kowal, Mozes Jacobs +22
What is the geometry of a visual percept? The most widely used protocols for decomposing neural network representations into interpretable parts treat concepts as isolated directio…
Interpretability as Compression: Reconsidering SAE Explanations of Neural Activations with MDL-SAEs
Kola Ayonrinde, Michael T. Pearce, Lee Sharkey
Sparse Autoencoders (SAEs) have emerged as a useful tool for interpreting the internal representations of neural networks. However, naively optimising SAEs for reconstruction loss…
Bilinear MLPs enable weight-based mechanistic interpretability
Michael T. Pearce, Thomas Dooms, Alice Rigg +2
A mechanistic understanding of how MLPs do computation in deep neural networks remains elusive. Current interpretability work can extract features from hidden activations over an i…
Weight-based Decomposition: A Case for Bilinear MLPs
Michael T. Pearce, Thomas Dooms, Alice Rigg
Gated Linear Units (GLUs) have become a common building block in modern foundation models. Bilinear layers drop the non-linearity in the "gate" but still have comparable performanc…