Pith. sign in

REVIEW 9 cited by

Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2411.01220 v2 pith:25WXSAZF submitted 2024-11-02 cs.LG

Enhancing Neural Network Interpretability with Feature-Aligned Sparse Autoencoders

classification cs.LG
keywords featuressaestextscdatainputinterpretabilitylearnlearned
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original 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.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. A Unifying Framework for Concept-Based Representational Similarity

    cs.LG 2026-06 unverdicted novelty 7.0

    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 s...

  2. Aligned Training: A Parameter-Free Method to Improve Feature Quality and Stability of Sparse Autoencoders (SAE)

    cs.LG 2026-05 unverdicted novelty 7.0

    Aligned training reparameterizes SAEs to enforce unit inner product between encoder and decoder directions, eliminating dead features and enhancing stability without hyperparameters.

  3. Improving Sparse Autoencoder with Dynamic Attention

    cs.LG 2026-04 unverdicted novelty 7.0

    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.

  4. Perplexity Can Miss SAE Feature Damage Under Quantization

    cs.LG 2026-06 unverdicted novelty 6.0

    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.

  5. Aligned Training: A Parameter-Free Method to Improve Feature Quality and Stability of Sparse Autoencoders (SAE)

    cs.LG 2026-05 unverdicted novelty 6.0

    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.

  6. Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates

    cs.CE 2026-03 unverdicted novelty 6.0

    Sparse autoencoders enable phase synchronization in frozen graph CFD surrogates through Hilbert-identified oscillatory features and SVD-based time-varying rotations.

  7. Query Circuits: Explaining How Language Models Answer User Prompts

    cs.AI 2025-09 conditional novelty 6.0

    Language models contain per-query circuits using a few percent of their edges that reproduce much of their answer behavior, and paraphrase-based Best-of-N search finds them.

  8. Cross-seed explainability using Procrustes-conditioned Joint End-to-end Top-K Sparse Autoencoders

    cs.CL 2026-07 conditional novelty 5.5

    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.

  9. Stable and Steerable Sparse Autoencoders with Weight Regularization

    stat.ML 2026-03 conditional novelty 5.0

    L2 weight regularization in TopK SAEs increases cross-seed feature overlap and roughly doubles measured steering success on Pythia-70M, at the cost of collapsing most latents to zero.