Toward Identifiable Sparse Autoencoders

• URL: http://arxiv.org/abs/2605.31245v1
• Date: Fri, 29 May 2026 12:44:45 GMT
• Title: Toward Identifiable Sparse Autoencoders
• Abstract summary: We show theoretically that sparse autoencoders (SAEs) are highly unstable.<n>We characterize the model properties that hinder the stability of real-world SAEs, and address each of these problems through minimal changes to the architecture and training procedure.<n>Together, these changes yield two versions of an bfidentifiable SAE (iSAE), a variant of the standard TopK SAE with lower reconstruction error and improved stability.
• Score: 28.407017043742783
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recently, sparse autoencoders (SAEs) have emerged as an attractive tool for interpreting and interacting with representations in practical neural networks. While it is common empirical folklore, we also show theoretically that SAEs are highly unstable: different training runs are likely to produce different concept dictionaries and sparse codes. We characterize the model properties that hinder the stability of real-world SAEs, and address each of these problems through minimal changes to the architecture and training procedure. Together, these changes yield two versions of an \textbf{i}dentifiable SAE (iSAE), a variant of the standard TopK SAE with lower reconstruction error and improved stability. We explain this improvement theoretically by connecting SAEs with traditional dictionary learning approaches, and show that the dictionaries learned in practice satisfy an approximate restricted isometry condition, rendering the corresponding sparse codes in those models near-identifiable.

Related papers

• Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performance [58.49623467782041]
  ELUDe is a method for improving the interpretability of deep neural networks.<n>It breaks latent representations into clear, inspectable sub-units that behave like interpretable features.<n>It requires no explicit training, no labels, and can be applied to pretrained models.
  arXiv  Detail & Related papers  (2026-05-29T13:39:09Z)
• Aligned Training: A Parameter-Free Method to Improve Feature Quality and Stability of Sparse Autoencoders (SAE) [2.406359246841227]
  Sparse autoencoders (SAEs) are used to interpret the inner workings of deep neural networks (DNNs)<n>SAEs exhibit critical shortcomings where a large fraction of features are never activated and are unstable.<n>We propose the textbfaligned training, a parameter-free re parameterization of SAEs that simultaneously improves reconstruction quality, eliminates dead features, and significantly enhances stability across training seeds.
  arXiv  Detail & Related papers  (2026-05-18T16:34:24Z)
• Taming Polysemanticity in LLMs: Provable Feature Recovery via Sparse Autoencoders [50.52694757593443]
  Existing SAE training algorithms often lack rigorous mathematical guarantees and suffer from practical limitations.<n>We first propose a novel statistical framework for the feature recovery problem, which includes a new notion of feature identifiability.<n>We introduce a new SAE training algorithm based on bias adaptation'', a technique that adaptively adjusts neural network bias parameters to ensure appropriate activation sparsity.
  arXiv  Detail & Related papers  (2025-06-16T20:58:05Z)
• Sparse Autoencoders, Again? [15.48801130346124]
  We formalize underappreciated weaknesses with both canonical SAEs and variational autoencoders.<n>We prove that global minima of our proposed model recover certain forms of structured data spread across a union of manifold.<n>In general, we are able to exceed the performance of equivalent-capacity SAEs and VAEs.
  arXiv  Detail & Related papers  (2025-06-05T10:26:06Z)
• Ensembling Sparse Autoencoders [10.81463830315253]
  Sparse autoencoders (SAEs) are used to decompose neural network activations into human-interpretable features.<n>We propose to ensemble multiple SAEs through naive bagging and boosting.<n>Our empirical results demonstrate that ensembling SAEs can improve the reconstruction of language model activations, diversity of features, and SAE stability.
  arXiv  Detail & Related papers  (2025-05-21T23:31:21Z)
• Tokenized SAEs: Disentangling SAE Reconstructions [0.9821874476902969]
  We show that RES-JB SAE features predominantly correspond to simple input statistics.<n>We propose a method that disentangles token reconstruction from feature reconstruction.
  arXiv  Detail & Related papers  (2025-02-24T17:04:24Z)
• Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models [16.894375498353092]
  Sparse Autoencoders (SAEs) have emerged as a powerful framework for machine learning interpretability.<n>Existing SAEs exhibit severe instability, as identical models trained on similar datasets can produce sharply different dictionaries.<n>We present Archetypal SAEs, wherein dictionary atoms are constrained to the convex hull of data.
  arXiv  Detail & Related papers  (2025-02-18T14:29:11Z)
• Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders [0.0]
  A recent line of work has shown promise in using sparse autoencoders (SAEs) to uncover interpretable features in neural network representations.<n>However, the simple linear-nonlinear encoding mechanism in SAEs limits their ability to perform accurate sparse inference.<n>We prove that an SAE encoder is inherently insufficient for accurate sparse inference, even in solvable cases.
  arXiv  Detail & Related papers  (2024-11-20T08:21:53Z)
• PseudoNeg-MAE: Self-Supervised Point Cloud Learning using Conditional Pseudo-Negative Embeddings [55.55445978692678]
  PseudoNeg-MAE enhances global feature representation of point cloud masked autoencoders by making them both discriminative and sensitive to transformations.<n>We propose a novel loss that explicitly penalizes invariant collapse, enabling the network to capture richer transformation cues while preserving discriminative representations.
  arXiv  Detail & Related papers  (2024-09-24T07:57:21Z)
• Generative error correction for code-switching speech recognition using large language models [49.06203730433107]
  Code-switching (CS) speech refers to the phenomenon of mixing two or more languages within the same sentence. We propose to leverage large language models (LLMs) and lists of hypotheses generated by an ASR to address the CS problem.
  arXiv  Detail & Related papers  (2023-10-17T14:49:48Z)
• Scalable Learning of Latent Language Structure With Logical Offline Cycle Consistency [71.42261918225773]
  Conceptually, LOCCO can be viewed as a form of self-learning where the semantic being trained is used to generate annotations for unlabeled text. As an added bonus, the annotations produced by LOCCO can be trivially repurposed to train a neural text generation model.
  arXiv  Detail & Related papers  (2023-05-31T16:47:20Z)
• SUN: Exploring Intrinsic Uncertainties in Text-to-SQL Parsers [61.48159785138462]
  This paper aims to improve the performance of text-to-dependence by exploring the intrinsic uncertainties in the neural network based approaches (called SUN) Extensive experiments on five benchmark datasets demonstrate that our method significantly outperforms competitors and achieves new state-of-the-art results.
  arXiv  Detail & Related papers  (2022-09-14T06:27:51Z)
• Unnatural Language Inference [48.45003475966808]
  We find that state-of-the-art NLI models, such as RoBERTa and BART, are invariant to, and sometimes even perform better on, examples with randomly reordered words. Our findings call into question the idea that our natural language understanding models, and the tasks used for measuring their progress, genuinely require a human-like understanding of syntax.
  arXiv  Detail & Related papers  (2020-12-30T20:40:48Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.