CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features

• URL: http://arxiv.org/abs/2508.12535v2
• Date: Fri, 17 Oct 2025 22:57:10 GMT
• Title: CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
• Authors: Seonglae Cho, Zekun Wu, Adriano Koshiyama,
• Abstract summary: We propose CorrSteer, which selects features by correlating sample correctness with SAE activations from generated tokens at inference time.<n>Our work establishes correlation-based selection as an effective and scalable approach for automated SAE steering across language model applications.
• Score: 1.5874067490843806
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse Autoencoders (SAEs) can extract interpretable features from large language models (LLMs) without supervision. However, their effectiveness in downstream steering tasks is limited by the requirement for contrastive datasets or large activation storage. To address these limitations, we propose CorrSteer, which selects features by correlating sample correctness with SAE activations from generated tokens at inference time. This approach uses only inference-time activations to extract more relevant features, thereby reducing spurious correlations. It also obtains steering coefficients from average activations, automating the entire pipeline. Our method shows improved task performance on QA, bias mitigation, jailbreaking prevention, and reasoning benchmarks on Gemma-2 2B and LLaMA-3.1 8B, notably achieving a +3.3% improvement in MMLU performance with 4000 samples and a +27.2% improvement in HarmBench with only 108 samples. Selected features demonstrate semantically meaningful patterns aligned with each task's requirements, revealing the underlying capabilities that drive performance. Our work establishes correlation-based selection as an effective and scalable approach for automated SAE steering across language model applications.

Related papers

• Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features [1.5874067490843806]
  Control Reinforcement Learning trains a policy to select SAE features for steering at each token, producing interpretable intervention logs.<n> Adaptive Feature Masking encourages diverse feature discovery while preserving singlefeature interpretability.<n>On Gemma 2 2B across MMLU, BBQ, GSM8K, HarmBench, and XSTest, CRL achieves improvements while providing per-token intervention logs.
  arXiv  Detail & Related papers  (2026-02-11T02:28:49Z)
• AceGRPO: Adaptive Curriculum Enhanced Group Relative Policy Optimization for Autonomous Machine Learning Engineering [52.67783579040657]
  AceGRPO is a machine learning system that prioritizes tasks at the agent's learning frontier to maximize learning efficiency.<n>Our trained Ace-30B model achieves a 100% valid submission rate on MLE-Bench-Lite, approaches the performance of proprietary frontier models, and outperforms larger open-source baselines.
  arXiv  Detail & Related papers  (2026-02-08T10:55:03Z)
• Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoders [63.544453925182005]
  We train 90 SAEs across three language models and evaluate their interpretability and steering utility.<n>Our analysis reveals only a relatively weak positive association (tau b approx 0.298), indicating that interpretability is an insufficient proxy for steering performance.<n>We propose a novel selection criterion called Delta Token Confidence, which measures how much amplifying a feature changes the next token distribution.
  arXiv  Detail & Related papers  (2025-10-04T04:14:50Z)
• GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs [56.93583799109029]
  GrAInS is an inference-time steering approach that operates across both language-only and vision-language models and tasks.<n>During inference, GrAInS hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale.<n>It consistently outperforms both fine-tuning and existing steering baselines.
  arXiv  Detail & Related papers  (2025-07-24T02:34:13Z)
• SPARE: Single-Pass Annotation with Reference-Guided Evaluation for Automatic Process Supervision and Reward Modelling [58.05959902776133]
  We introduce Single-Pass.<n>with Reference-Guided Evaluation (SPARE), a novel structured framework that enables efficient per-step annotation.<n>We demonstrate SPARE's effectiveness across four diverse datasets spanning mathematical reasoning (GSM8K, MATH), multi-hop question answering (MuSiQue-Ans), and spatial reasoning (SpaRP)<n>On ProcessBench, SPARE demonstrates data-efficient out-of-distribution generalization, using only $sim$16% of training samples compared to human-labeled and other synthetically trained baselines.
  arXiv  Detail & Related papers  (2025-06-18T14:37:59Z)
• Fusion Steering: Prompt-Specific Activation Control [0.0]
  Fusion Steering improves factual accuracy in large language models (LLMs) for question-answering (QA) tasks.<n>This approach introduces flexible steering configurations, including full-layer steering and segmented steering.<n>Under the stricter SimpleQA rubric, segmented steering boosts fully correct responses from 0.0% to 13.1%.
  arXiv  Detail & Related papers  (2025-05-28T16:46:55Z)
• R-Sparse: Rank-Aware Activation Sparsity for Efficient LLM Inference [77.47238561728459]
  R-Sparse is a training-free activation sparsity approach capable of achieving high sparsity levels in advanced LLMs.<n> Experiments on Llama-2/3 and Mistral models across ten diverse tasks demonstrate that R-Sparse achieves comparable performance at 50% model-level sparsity.
  arXiv  Detail & Related papers  (2025-04-28T03:30:32Z)
• AskToAct: Enhancing LLMs Tool Use via Self-Correcting Clarification [25.27444694706659]
  We present AskToAct, which exploits the structural mapping between queries and their tool invocation solutions.<n>By systematically removing key parameters from queries while retaining them as ground truth, we enable automated construction of high-quality training data.<n>Our framework exhibits robust performance across different model architectures and successfully generalizes to entirely unseen APIs without additional training.
  arXiv  Detail & Related papers  (2025-03-03T12:55:49Z)
• Multi-Attribute Steering of Language Models via Targeted Intervention [56.93583799109029]
  Inference-time intervention (ITI) has emerged as a promising method for steering large language model (LLM) behavior in a particular direction.<n>We introduce Multi-Attribute Targeted Steering (MAT-Steer), a novel steering framework designed for selective token-level intervention across multiple attributes.
  arXiv  Detail & Related papers  (2025-02-18T02:27:23Z)
• PIVOT-R: Primitive-Driven Waypoint-Aware World Model for Robotic Manipulation [68.17081518640934]
  We propose a PrIrmitive-driVen waypOinT-aware world model for Robotic manipulation (PIVOT-R) PIVOT-R consists of a Waypoint-aware World Model (WAWM) and a lightweight action prediction module. Our PIVOT-R outperforms state-of-the-art open-source models on the SeaWave benchmark, achieving an average relative improvement of 19.45% across four levels of instruction tasks.
  arXiv  Detail & Related papers  (2024-10-14T11:30:18Z)
• Efficiently Deploying LLMs with Controlled Risk [0.9208007322096532]
  We present hierarchical chains with multi-level abstention (HCMA), which use model-intrinsic uncertainty to delegate queries. Our framework presents novel trade-offs between efficiency and risk.
  arXiv  Detail & Related papers  (2024-10-03T03:25:56Z)
• Get my drift? Catching LLM Task Drift with Activation Deltas [55.75645403965326]
  Task drift allows attackers to exfiltrate data or influence the LLM's output for other users.<n>We show that a simple linear classifier can detect drift with near-perfect ROC AUC on an out-of-distribution test set.<n>We observe that this approach generalizes surprisingly well to unseen task domains, such as prompt injections, jailbreaks, and malicious instructions.
  arXiv  Detail & Related papers  (2024-06-02T16:53:21Z)
• Localizing Task Information for Improved Model Merging and Compression [61.16012721460561]
  We show that the information required to solve each task is still preserved after merging as different tasks mostly use non-overlapping sets of weights. We propose Consensus Merging, an algorithm that eliminates such weights and improves the general performance of existing model merging approaches.
  arXiv  Detail & Related papers  (2024-05-13T14:54:37Z)
• Robusta: Robust AutoML for Feature Selection via Reinforcement Learning [24.24652530951966]
  We propose the first robust AutoML framework, Robusta--based on reinforcement learning (RL) We show that the framework is able to improve the model robustness by up to 22% while maintaining competitive accuracy on benign samples.
  arXiv  Detail & Related papers  (2021-01-15T03:12:29Z)

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.