Evaluating Sparse Autoencoders for Monosemantic Representation

• URL: http://arxiv.org/abs/2508.15094v1
• Date: Wed, 20 Aug 2025 22:08:01 GMT
• Title: Evaluating Sparse Autoencoders for Monosemantic Representation
• Authors: Moghis Fereidouni, Muhammad Umair Haider, Peizhong Ju, A. B. Siddique,
• Abstract summary: A key barrier to interpreting large language models is polysemanticity, where neurons activate for multiple unrelated concepts.<n>Sparse autoencoders (SAEs) have been proposed to mitigate this issue by transforming dense activations into sparse, more interpretable features.<n>This paper provides the first systematic evaluation of SAEs against base models concerning monosemanticity.
• Score: 7.46972338257749
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: A key barrier to interpreting large language models is polysemanticity, where neurons activate for multiple unrelated concepts. Sparse autoencoders (SAEs) have been proposed to mitigate this issue by transforming dense activations into sparse, more interpretable features. While prior work suggests that SAEs promote monosemanticity, there has been no quantitative comparison with their base models. This paper provides the first systematic evaluation of SAEs against base models concerning monosemanticity. We introduce a fine-grained concept separability score based on the Jensen-Shannon distance, which captures how distinctly a neuron's activation distributions vary across concepts. Using Gemma-2-2B and multiple SAE variants across five benchmarks, we show that SAEs reduce polysemanticity and achieve higher concept separability. However, greater sparsity of SAEs does not always yield better separability and often impairs downstream performance. To assess practical utility, we evaluate concept-level interventions using two strategies: full neuron masking and partial suppression. We find that, compared to base models, SAEs enable more precise concept-level control when using partial suppression. Building on this, we propose Attenuation via Posterior Probabilities (APP), a new intervention method that uses concept-conditioned activation distributions for targeted suppression. APP outperforms existing approaches in targeted concept removal.

Related papers

• Learning a Generative Meta-Model of LLM Activations [75.30161960337892]
  We create "meta-models" that learn the distribution of a network's internal states.<n>Applying the meta-model's learned prior to steering interventions improves fluency, with larger gains as loss decreases.<n>These results suggest generative meta-models offer a scalable path toward interpretability without restrictive structural assumptions.
  arXiv  Detail & Related papers  (2026-02-06T18:59:56Z)
• Concept Component Analysis: A Principled Approach for Concept Extraction in LLMs [51.378834857406325]
  Mechanistic interpretability seeks to mitigate the issues through extracts from large language models.<n>Sparse autoencoders (SAEs) have emerged as a popular approach for extracting interpretable and monosemantic concepts.<n>We show that SAEs suffer from a fundamental theoretical ambiguity: the well-defined correspondence between LLM representations and human-interpretable concepts remains unclear.
  arXiv  Detail & Related papers  (2026-01-28T09:27:05Z)
• CASL: Concept-Aligned Sparse Latents for Interpreting Diffusion Models [45.90361318326864]
  Internal activations of diffusion models encode rich semantic information, but interpreting such representations remains challenging.<n>We introduce CASL (Concept-Aligned Sparse Latents), a supervised framework that aligns sparse latent dimensions of diffusion models with semantic concepts.<n>Unlike editing methods, CASL-Steer is used solely as a causal probe to reveal how concept-aligned latents influence generated content.
  arXiv  Detail & Related papers  (2026-01-21T20:14:17Z)
• AbsTopK: Rethinking Sparse Autoencoders For Bidirectional Features [19.58274892471746]
  Sparse autoencoders (SAEs) have emerged as powerful techniques for interpretability of large language models.<n>We introduce such a framework by unrolling the proximal gradient method for sparse coding.<n>We show that a single-step update naturally recovers common SAE variants, including ReLU, JumpReLU, and TopK.
  arXiv  Detail & Related papers  (2025-10-01T01:29:31Z)
• SAEmnesia: Erasing Concepts in Diffusion Models with Sparse Autoencoders [6.6477077425454745]
  SAEmnesia is a supervised sparse autoencoder training method that promotes one-to-one concept-neuron mappings.<n>Our approach learns specialized neurons with significantly stronger concept associations compared to unsupervised baselines.
  arXiv  Detail & Related papers  (2025-09-23T11:29:30Z)
• On the Theoretical Understanding of Identifiable Sparse Autoencoders and Beyond [36.107366496809675]
  Sparse autoencoders (SAEs) have emerged as a powerful tool for interpreting features learned by large language models (LLMs)<n>It aims to recover complex superposed polysemantic features into interpretable monosemantic ones through feature reconstruction via sparsely activated neural networks.<n>Despite the wide applications of SAEs, it remains unclear under what conditions an SAE can fully recover the ground truth monosemantic features from the superposed polysemantic ones.
  arXiv  Detail & Related papers  (2025-06-19T02:16:08Z)
• 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)
• Decomposing MLP Activations into Interpretable Features via Semi-Nonnegative Matrix Factorization [17.101290138120564]
  Current methods rely on dictionary learning with sparse autoencoders (SAEs)<n>Here, we tackle these limitations by directly decomposing activations with semi-nonnegative matrix factorization (SNMF)<n>Experiments on Llama 3.1, Gemma 2 and GPT-2 show that SNMF derived features outperform SAEs and a strong supervised baseline (difference-in-means) on causal steering.
  arXiv  Detail & Related papers  (2025-06-12T17:33:29Z)
• Interpretable Few-Shot Image Classification via Prototypical Concept-Guided Mixture of LoRA Experts [79.18608192761512]
  Self-Explainable Models (SEMs) rely on Prototypical Concept Learning (PCL) to enable their visual recognition processes more interpretable.<n>We propose a Few-Shot Prototypical Concept Classification framework that mitigates two key challenges under low-data regimes: parametric imbalance and representation misalignment.<n>Our approach consistently outperforms existing SEMs by a notable margin, with 4.2%-8.7% relative gains in 5-way 5-shot classification.
  arXiv  Detail & Related papers  (2025-06-05T06:39:43Z)
• 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)
• SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoders [4.013156524547073]
  Diffusion models can inadvertently generate harmful or undesirable content.<n>Recent machine unlearning approaches offer potential solutions but often lack transparency.<n>We introduce SAeUron, a novel method leveraging features learned by sparse autoencoders to remove unwanted concepts.
  arXiv  Detail & Related papers  (2025-01-29T23:29:47Z)
• Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness [68.69369585600698]
  Deep learning models often suffer from a lack of interpretability due to polysemanticity. Recent advances in monosemanticity, where neurons correspond to consistent and distinct semantics, have significantly improved interpretability. We show that monosemantic features not only enhance interpretability but also bring concrete gains in model performance.
  arXiv  Detail & Related papers  (2024-10-27T18:03:20Z)
• Sparse Prototype Network for Explainable Pedestrian Behavior Prediction [60.80524827122901]
  We present Sparse Prototype Network (SPN), an explainable method designed to simultaneously predict a pedestrian's future action, trajectory, and pose. Regularized by mono-semanticity and clustering constraints, the prototypes learn consistent and human-understandable features.
  arXiv  Detail & Related papers  (2024-10-16T03:33:40Z)
• MulCPred: Learning Multi-modal Concepts for Explainable Pedestrian Action Prediction [57.483718822429346]
  MulCPred is proposed that explains its predictions based on multi-modal concepts represented by training samples. MulCPred is evaluated on multiple datasets and tasks.
  arXiv  Detail & Related papers  (2024-09-14T14:15:28Z)
• Regularizing Variational Autoencoder with Diversity and Uncertainty Awareness [61.827054365139645]
  Variational Autoencoder (VAE) approximates the posterior of latent variables based on amortized variational inference. We propose an alternative model, DU-VAE, for learning a more Diverse and less Uncertain latent space.
  arXiv  Detail & Related papers  (2021-10-24T07:58:13Z)

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.