A unifying framework decomposes concept alignment into instance-wise and distributional translation and concept consistency, introduces the InterVenchA benchmark, and shows that joint optimization via CoSAE recovers strong alignment even with 0.1% paired data.
Enhancing neural network interpretability with feature-aligned sparse autoencoders
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
abstract
Sparse Autoencoders (SAEs) have shown promise in improving the interpretability of neural network activations, but can learn features that are not features of the input, limiting their effectiveness. We propose \textsc{Mutual Feature Regularization} \textbf{(MFR)}, a regularization technique for improving feature learning by encouraging SAEs trained in parallel to learn similar features. We motivate \textsc{MFR} by showing that features learned by multiple SAEs are more likely to correlate with features of the input. By training on synthetic data with known features of the input, we show that \textsc{MFR} can help SAEs learn those features, as we can directly compare the features learned by the SAE with the input features for the synthetic data. We then scale \textsc{MFR} to SAEs that are trained to denoise electroencephalography (EEG) data and SAEs that are trained to reconstruct GPT-2 Small activations. We show that \textsc{MFR} can improve the reconstruction loss of SAEs by up to 21.21\% on GPT-2 Small, and 6.67\% on EEG data. Our results suggest that the similarity between features learned by different SAEs can be leveraged to improve SAE training, thereby enhancing performance and the usefulness of SAEs for model interpretability.
years
2026 6representative citing papers
A cross-attention SAE with sparsemax attention achieves lower reconstruction loss and higher-quality concepts than fixed-sparsity baselines by making activation counts data-dependent.
Quantization of LLMs can degrade many SAE features even when perplexity improves or stays similar, as shown by correlation measurements on frozen SAEs for Pythia-70M and Gemma-2-2B models across INT8 to INT4.
Aligned training reparameterizes SAEs to enforce unit alignment between encoder and decoder directions, yielding Pareto gains on SAEBench while removing dead features and improving stability.
Sparse autoencoders enable phase synchronization in frozen graph CFD surrogates through Hilbert-identified oscillatory features and SVD-based time-varying rotations.
Procrustes-conditioned joint End-to-end Top-K SAEs recover more cross-seed universal features (r≥0.70) from independent BERT seeds than post-hoc alignment baselines on three NLP datasets.
citing papers explorer
-
A Unifying Framework for Concept-Based Representational Similarity
A unifying framework decomposes concept alignment into instance-wise and distributional translation and concept consistency, introduces the InterVenchA benchmark, and shows that joint optimization via CoSAE recovers strong alignment even with 0.1% paired data.
-
Improving Sparse Autoencoder with Dynamic Attention
A cross-attention SAE with sparsemax attention achieves lower reconstruction loss and higher-quality concepts than fixed-sparsity baselines by making activation counts data-dependent.
-
Perplexity Can Miss SAE Feature Damage Under Quantization
Quantization of LLMs can degrade many SAE features even when perplexity improves or stays similar, as shown by correlation measurements on frozen SAEs for Pythia-70M and Gemma-2-2B models across INT8 to INT4.
-
Aligned Training: A Parameter-Free Method to Improve Feature Quality and Stability of Sparse Autoencoders (SAE)
Aligned training reparameterizes SAEs to enforce unit alignment between encoder and decoder directions, yielding Pareto gains on SAEBench while removing dead features and improving stability.
-
Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates
Sparse autoencoders enable phase synchronization in frozen graph CFD surrogates through Hilbert-identified oscillatory features and SVD-based time-varying rotations.
-
Cross-seed explainability using Procrustes-conditioned Joint End-to-end Top-K Sparse Autoencoders
Procrustes-conditioned joint End-to-end Top-K SAEs recover more cross-seed universal features (r≥0.70) from independent BERT seeds than post-hoc alignment baselines on three NLP datasets.