Feature Hedging: Correlated Features Break Narrow Sparse Autoencoders

• URL: http://arxiv.org/abs/2505.11756v1
• Date: Fri, 16 May 2025 23:30:17 GMT
• Title: Feature Hedging: Correlated Features Break Narrow Sparse Autoencoders
• Authors: David Chanin, Tomáš Dulka, Adrià Garriga-Alonso,
• Abstract summary: It is assumed that sparse autoencoders (SAEs) decompose polysemantic activations into interpretable linear directions.<n>We find that if an SAE is more narrow than the number of underlying "true features" on which it is trained, and there is correlation between features, the SAE will merge components of correlated features together.<n>This phenomenon, which we call feature hedging, is caused by SAE reconstruction loss, and is more severe the narrower the SAE.
• Score: 1.0582505915332336
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: It is assumed that sparse autoencoders (SAEs) decompose polysemantic activations into interpretable linear directions, as long as the activations are composed of sparse linear combinations of underlying features. However, we find that if an SAE is more narrow than the number of underlying "true features" on which it is trained, and there is correlation between features, the SAE will merge components of correlated features together, thus destroying monosemanticity. In LLM SAEs, these two conditions are almost certainly true. This phenomenon, which we call feature hedging, is caused by SAE reconstruction loss, and is more severe the narrower the SAE. In this work, we introduce the problem of feature hedging and study it both theoretically in toy models and empirically in SAEs trained on LLMs. We suspect that feature hedging may be one of the core reasons that SAEs consistently underperform supervised baselines. Finally, we use our understanding of feature hedging to propose an improved variant of matryoshka SAEs. Our work shows there remain fundamental issues with SAEs, but we are hopeful that that highlighting feature hedging will catalyze future advances that allow SAEs to achieve their full potential of interpreting LLMs at scale.

Related papers

• Dense SAE Latents Are Features, Not Bugs [75.08462524662072]
  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)
• 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)
• Factual Self-Awareness in Language Models: Representation, Robustness, and Scaling [56.26834106704781]
  Factual incorrectness in generated content is one of the primary concerns in ubiquitous deployment of large language models (LLMs)<n>We provide evidence supporting the presence of LLMs' internal compass that dictate the correctness of factual recall at the time of generation.<n>Scaling experiments across model sizes and training dynamics highlight that self-awareness emerges rapidly during training and peaks in intermediate layers.
  arXiv  Detail & Related papers  (2025-05-27T16:24:02Z)
• 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)
• Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models [48.40096116617163]
  Large Language Models (LLMs) demonstrate the ability to solve reasoning and mathematical problems using the Chain-of-Thought (CoT) technique.<n>This work, inspired by the deep thinking paradigm of DeepSeek-R1, utilizes a steering technique to enhance the reasoning ability of an LLM without external datasets.
  arXiv  Detail & Related papers  (2025-05-21T15:17:59Z)
• Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models [50.587868616659826]
  Sparse Autoencoders (SAEs) have been shown to enhance interpretability and steerability in Large Language Models (LLMs)<n>In this work, we extend the application of SAEs to Vision-Language Models (VLMs), such as CLIP, and introduce a comprehensive framework for evaluating monosemanticity in vision representations.
  arXiv  Detail & Related papers  (2025-04-03T17:58:35Z)
• Evaluating and Designing Sparse Autoencoders by Approximating Quasi-Orthogonality [3.9230690073443166]
  We show that the magnitude of sparse feature vectors can be approximated using their corresponding dense vector with a closed-form error bound.<n>We introduce Approximate Activation Feature (AFA), which approximates the magnitude of the ground-truth sparse feature vector.<n>We demonstrate that top-AFA SAEs achieve reconstruction loss comparable to that of state-of-the-art top-k SAEs.
  arXiv  Detail & Related papers  (2025-03-31T16:22:11Z)
• Are Sparse Autoencoders Useful? A Case Study in Sparse Probing [6.836374436707495]
  Sparse autoencoders (SAEs) are a popular method for interpreting concepts represented in large language model (LLM) activations.<n>One alternative source of evidence would be demonstrating that SAEs improve performance on downstream tasks beyond existing baselines.<n>We test this by applying SAEs to the real-world task of LLM activation probing in four regimes.
  arXiv  Detail & Related papers  (2025-02-23T18:54:15Z)
• Sparse Autoencoders Trained on the Same Data Learn Different Features [0.7234862895932991]
  Sparse autoencoders (SAEs) are a useful tool for uncovering human-interpretable features in large language models.<n>Our research shows that SAEs trained on the same model and data, differing only in the random seed used to initialize their weights, identify different sets of features.
  arXiv  Detail & Related papers  (2025-01-28T01:24:16Z)
• Adaptive Pruning for Large Language Models with Structural Importance Awareness [66.2690963378878]
  Large language models (LLMs) have significantly improved language understanding and generation capabilities.<n>LLMs are difficult to deploy on resource-constrained edge devices due to their high computational and storage resource demands.<n>We propose structurally-aware adaptive pruning (SAAP) to significantly reduce the computational and memory costs while maintaining model performance.
  arXiv  Detail & Related papers  (2024-12-19T18:08:04Z)
• Understanding the Role of Equivariance in Self-supervised Learning [51.56331245499712]
  equivariant self-supervised learning (E-SSL) learns features to be augmentation-aware. We identify a critical explaining-away effect in E-SSL that creates a synergy between the equivariant and classification tasks. We reveal several principles for practical designs of E-SSL.
  arXiv  Detail & Related papers  (2024-11-10T16:09:47Z)
• Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders [115.34050914216665]
  Sparse Autoencoders (SAEs) have emerged as a powerful unsupervised method for extracting sparse representations from language models. We introduce a suite of 256 SAEs, trained on each layer and sublayer of the Llama-3.1-8B-Base model, with 32K and 128K features. We assess the generalizability of SAEs trained on base models to longer contexts and fine-tuned models.
  arXiv  Detail & Related papers  (2024-10-27T17:33:49Z)
• Interpretability as Compression: Reconsidering SAE Explanations of Neural Activations with MDL-SAEs [0.0]
  We present an information-theoretic framework for interpreting SAEs as lossy compression algorithms. We argue that using MDL rather than sparsity may avoid potential pitfalls with naively maximising sparsity.
  arXiv  Detail & Related papers  (2024-10-15T01:38:03Z)
• A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders [0.0]
  We show that sparse decomposition and splitting of hierarchical features is not robust.<n>Specifically, we show that seemingly monosemantic features fail to fire where they should, and instead get "absorbed" into their children features.
  arXiv  Detail & Related papers  (2024-09-22T16:11:02Z)
• 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)
• The Strong Pull of Prior Knowledge in Large Language Models and Its Impact on Emotion Recognition [74.04775677110179]
  In-context Learning (ICL) has emerged as a powerful paradigm for performing natural language tasks with Large Language Models (LLM) We show that LLMs have strong yet inconsistent priors in emotion recognition that ossify their predictions. Our results suggest that caution is needed when using ICL with larger LLMs for affect-centered tasks outside their pre-training domain.
  arXiv  Detail & Related papers  (2024-03-25T19:07:32Z)

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.