Evaluating Sparse Autoencoders: From Shallow Design to Matching Pursuit
This work addresses a bottleneck in interpretability tools for neural networks, offering an incremental improvement for researchers in AI interpretability.
The paper tackled the limitation of shallow sparse autoencoders in extracting correlated features by proposing an iterative Matching Pursuit-based SAE, which improved feature extraction in hierarchical settings like MNIST digit generation with guaranteed monotonic reconstruction improvement.
Sparse autoencoders (SAEs) have recently become central tools for interpretability, leveraging dictionary learning principles to extract sparse, interpretable features from neural representations whose underlying structure is typically unknown. This paper evaluates SAEs in a controlled setting using MNIST, which reveals that current shallow architectures implicitly rely on a quasi-orthogonality assumption that limits the ability to extract correlated features. To move beyond this, we compare them with an iterative SAE that unrolls Matching Pursuit (MP-SAE), enabling the residual-guided extraction of correlated features that arise in hierarchical settings such as handwritten digit generation while guaranteeing monotonic improvement of the reconstruction as more atoms are selected.