mlsanityData to 14 Oct 2026

Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders

arXiv 2411.13117 ↗First posted 20 Nov 2024cs.LGcited by 18

On the map: Mechanistic interpretability → Sparse autoencoder training and evaluation

Abstract (arXiv)

A recent line of work has shown promise in using sparse autoencoders (SAEs) to uncover interpretable features in neural network representations. However, the simple linear-nonlinear encoding mechanism in SAEs limits their ability to perform accurate sparse inference. Using compressed sensing theory, we prove that an SAE encoder is inherently insufficient for accurate sparse inference, even in solvable cases. We then decouple encoding and decoding processes to empirically explore conditions where more sophisticated sparse inference methods outperform traditional SAE encoders. Our results reveal substantial performance gains with minimal compute increases in correct inference of sparse codes. We demonstrate this generalises to SAEs applied to large language models, where more expressive encoders achieve greater interpretability. This work opens new avenues for understanding neural network representations and analysing large language model activations.

Key facts. Cited by 18 papers in the map (data through 14 Oct 2026). Placed in Sparse autoencoder training and evaluation, one of 13 directions in Mechanistic interpretability. Builds on 15 of its 38 references that are in the map. First posted on arXiv 20 Nov 2024.

Builds on

15 of 38 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 29 Aug
  2. 2024Interpreting Attention Layer Outputs with Sparse Autoencoders25 Jun
  3. 2024The Missing Curve Detectors of InceptionV1: Applying Sparse Autoencoders to InceptionV1 Early Vision6 Jun
  4. 2024Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control14 May
  5. 2022Toy Models of Superposition21 Sep

Other references · 10

  1. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
  2. 2024Scaling and evaluating sparse autoencoderscited by 553
  3. 2024Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoderscited by 300
  4. 2013k-Sparse Autoencoderscited by 215
  5. 2024Improving Dictionary Learning with Gated Sparse Autoencoderscited by 179
  6. 2022Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networkscited by 129
  7. 2018Iterative Amortized Inferencecited by 94
  8. 2020On Interpretability of Artificial Neural Networks: A Surveycited by 88
  9. 2021A Survey on Understanding, Visualizations, and Explanation of Deep Neural Networkscited by 14
  10. 2023Identifying Interpretable Visual Features in Artificial and Biological Neural Systemscited by 9

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 9; 2026: 9.

18 papers.

Sort:

By area:

  1. 2025From superposition to sparse codes: interpretable representations in neural networkscited by 16
  2. 2025Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEscited by 14
  3. 2026SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Datacited by 11
  4. 2025AbsTopK: Rethinking Sparse Autoencoders For Bidirectional Featurescited by 9
  5. 2026Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisationcited by 6
  6. 2025Superposition disentanglement of neural representations reveals hidden alignmentcited by 4
  7. 2026Are Sparse Autoencoder Benchmarks Reliable?cited by 4
  8. 2025Resurrecting the Salmon: Rethinking Mechanistic Interpretability with Domain-Specific Sparse Autoencoderscited by 3
  9. 2026The Rate-Distortion-Polysemanticity Tradeoff in SAEscited by 3
  10. 2025MedSAE: Dissecting MedCLIP Representations with Sparse Autoencoderscited by 2
  11. 2025Kronecker Factorization Improves Efficiency and Interpretability of Sparse Autoencoderscited by 0
  12. 2025Towards Atoms of Large Language Modelscited by 0
6 more
  1. 2025Equivariant Sparse Autoencoders: Mechanistic Interpretability of Neural Networks on Symmetric Datacited by 0
  2. 2026Similarity of Neural Network Representations in Superpositioncited by 0
  3. 2026Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activationscited by 0
  4. 2026How Optimality Structures Sparse Dictionaries: Theory for Interpreting SAE Representationscited by 0
  5. 2026Interference Beyond Geometry in Concept Extractioncited by 0
  6. 2026The Independence Prior of SAEs Fragments Visual Conceptscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025On the transferability of Sparse Autoencoders for interpreting compressed models
  2. 2026Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability
  3. 2026Inference and learning in sparse autoencoders as natural gradient flow
  4. 2025Empirical Evaluation of Progressive Coding for Sparse Autoencoders
  5. 2026Toward Identifiable Sparse Autoencoders
  6. 2026Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?
  7. 2024Interpretability as Compression: Reconsidering SAE Explanations of Neural Activations with MDL-SAEs
  8. 2025SplInterp: Improving our Understanding and Training of Sparse Autoencoders