Enforcing Orderedness to Improve Feature Consistency
This addresses feature inconsistency in neural network interpretability, though it appears incremental as it builds on prior nested SAE methods.
The paper tackled the problem of inconsistent learned features in sparse autoencoders by introducing Ordered Sparse Autoencoders (OSAE), which enforce strict ordering and deterministic use of latent features, showing improved consistency on models like Gemma2-2B and Pythia-70M compared to baselines.
Sparse autoencoders (SAEs) have been widely used for interpretability of neural networks, but their learned features often vary across seeds and hyperparameter settings. We introduce Ordered Sparse Autoencoders (OSAE), which extend Matryoshka SAEs by (1) establishing a strict ordering of latent features and (2) deterministically using every feature dimension, avoiding the sampling-based approximations of prior nested SAE methods. Theoretically, we show that OSAEs resolve permutation non-identifiability in settings of sparse dictionary learning where solutions are unique (up to natural symmetries). Empirically on Gemma2-2B and Pythia-70M, we show that OSAEs can help improve consistency compared to Matryoshka baselines.