From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability

• URL: http://arxiv.org/abs/2606.18383v1
• Date: Tue, 16 Jun 2026 18:28:23 GMT
• Title: From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability
• Abstract summary: We study when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM.<n>Our framework derives an upper bound on the base model's expected risk using four measurable quantities.<n>We show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes.
• Score: 53.681678236115836
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction. Our framework derives an upper bound on the base model's expected risk using four measurable quantities: proxy risk, SAE reconstruction gap, concept-pool mismatch, and sparse complexity. We interpret this certificate as an operational criterion for explanatory faithfulness. In particular, a non-vacuous bound indicates that the extracted sparse features retain meaningful predictive information, while small reconstruction and mismatch errors indicate that the proxy remains behaviorally close to the original model. Empirically, we show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes. A detailed layerwise analysis of Llama-3-8B reveals a strong depth dependence, with later layers becoming much easier to certify, associated with both stronger local fidelity and weaker downstream error amplification. Finally, through feature-shuffling ablations, we show that the decomposition distinguishes genuine semantic alignment from mere statistical sparsity, providing a useful diagnostic for when SAE-based explanations become less reliable.

Related papers

• When Pruning Meets Interpretability: Preserving Sparse Autoencoder Robustness in LLMs [15.418870425179875]
  Sparse autoencoders (SAEs) are widely used to interpret the internal representations of large language models.<n>We show that for a fixed SAE, its impact is governed by perturbation energy.<n>We propose a layer-wise sparsity allocation strategy, achieving lower perplexity under the same average pruning sparsity.
  arXiv  Detail & Related papers  (2026-08-26T15:57:42Z)
• Quantifying Prior Dominance in RAG Systems [1.9544213396776273]
  We introduce the Normalized Context Utilization (NCU) metric to quantify contextual information gain.<n>We evaluate architectures ranging from 1.5B to 72B parameters alongside a proprietary commercial API.<n>Our findings highlight the structural advantage and superior contextual adherence of SLMs in strict extraction.
  arXiv  Detail & Related papers  (2026-04-29T15:38:24Z)
• Sharp Convergence Rates for Masked Diffusion Models [53.117058231393834]
  We develop a total-variation based analysis for the Euler method that overcomes limitations.<n>Our results relax assumptions on score estimation, improve parameter dependencies, and establish convergence guarantees.<n>Overall, our analysis introduces a direct TV-based error decomposition along the CTMC trajectory and a decoupling-based path-wise analysis for FHS.
  arXiv  Detail & Related papers  (2026-02-26T00:47:51Z)
• Sparse Semantic Dimension as a Generalization Certificate for LLMs [53.681678236115836]
  We introduce the Sparse Semantic Dimension (SSD), a complexity measure derived from the active feature vocabulary of a Sparse Autoencoder (SAE) trained on the model's layers.<n>We validate this framework on GPT-2 Small and Gemma-2B, demonstrating that our bound provides non-vacuous certificates at realistic sample sizes.
  arXiv  Detail & Related papers  (2026-02-11T21:45:18Z)
• Analysis of Variational Sparse Autoencoders [1.675385127117872]
  We investigate whether incorporating variational methods into SAE architectures can improve feature organization and interpretability.<n>We introduce the Variational Sparse Autoencoder (vSAE), which replaces deterministic ReLU gating with sampling from learned Gaussian posteriors.<n>Our findings suggest that naive application of variational methods to SAEs does not improve feature organization or interpretability.
  arXiv  Detail & Related papers  (2025-09-26T23:09:56Z)
• Resurrecting the Salmon: Rethinking Mechanistic Interpretability with Domain-Specific Sparse Autoencoders [0.0]
  We show that restricting SAE training to a well-defined domain reallocates capacity to domain-specific features.<n>We find that domain-confined SAEs explain up to 20% more variance, achieve higher loss recovery, and reduce linear residual error.
  arXiv  Detail & Related papers  (2025-08-12T21:45:10Z)
• 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)
• Dense SAE Latents Are Features, Not Bugs [86.50389855919292]
  We show that dense latents serve functional roles in language model computation.<n>We identify classes tied to position tracking, context binding, entropy regulation, letter-specific output signals, part-of-speech, and principal component reconstruction.
  arXiv  Detail & Related papers  (2025-06-18T17:59:35Z)
• Evaluating and Designing Sparse Autoencoders by Approximating Quasi-Orthogonality [3.9230690073443166]
  We introduce a novel activation function, top-AFA, which builds upon our formulation of approximate feature activation (AFA)<n>By training SAEs on three intermediate layers to reconstruct GPT2 hidden embeddings for over 80 million tokens from the OpenWebText dataset, we demonstrate the empirical merits of this approach.
  arXiv  Detail & Related papers  (2025-03-31T16:22:11Z)
• Unconditional Truthfulness: Learning Unconditional Uncertainty of Large Language Models [104.55763564037831]
  We train a regression model that leverages attention maps, probabilities on the current generation step, and recurrently computed uncertainty scores from previously generated tokens.<n>Our evaluation shows that the proposed method is highly effective for selective generation, achieving substantial improvements over rivaling unsupervised and supervised approaches.
  arXiv  Detail & Related papers  (2024-08-20T09:42:26Z)
• Cycles of Thought: Measuring LLM Confidence through Stable Explanations [53.15438489398938]
  Large language models (LLMs) can reach and even surpass human-level accuracy on a variety of benchmarks, but their overconfidence in incorrect responses is still a well-documented failure mode. We propose a framework for measuring an LLM's uncertainty with respect to the distribution of generated explanations for an answer.
  arXiv  Detail & Related papers  (2024-06-05T16:35:30Z)

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.