Beyond Linear Steering: Unified Multi-Attribute Control for Language Models

• URL: http://arxiv.org/abs/2505.24535v1
• Date: Fri, 30 May 2025 12:41:19 GMT
• Title: Beyond Linear Steering: Unified Multi-Attribute Control for Language Models
• Authors: Narmeen Oozeer, Luke Marks, Fazl Barez, Amirali Abdullah,
• Abstract summary: We introduce K-Steering, a unified and flexible approach that trains a single non-linear multi-label classifier on hidden activations.<n>This avoids linearity assumptions, removes the need for storing and tuning separate vectors attribute, and allows dynamic composition of behaviors without retraining.<n> Empirical results across 3 model families, validated by both activation-based classifiers and LLM-based judges, demonstrate that K-Steering outperforms strong baselines in accurately steering multiple behaviors.
• Score: 1.6874375111244329
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Controlling multiple behavioral attributes in large language models (LLMs) at inference time is a challenging problem due to interference between attributes and the limitations of linear steering methods, which assume additive behavior in activation space and require per-attribute tuning. We introduce K-Steering, a unified and flexible approach that trains a single non-linear multi-label classifier on hidden activations and computes intervention directions via gradients at inference time. This avoids linearity assumptions, removes the need for storing and tuning separate attribute vectors, and allows dynamic composition of behaviors without retraining. To evaluate our method, we propose two new benchmarks, ToneBank and DebateMix, targeting compositional behavioral control. Empirical results across 3 model families, validated by both activation-based classifiers and LLM-based judges, demonstrate that K-Steering outperforms strong baselines in accurately steering multiple behaviors.

Related papers

• Unified modality separation: A vision-language framework for unsupervised domain adaptation [60.8391821117794]
  Unsupervised domain adaptation (UDA) enables models trained on a labeled source domain to handle new unlabeled domains.<n>We propose a unified modality separation framework that accommodates both modality-specific and modality-invariant components.<n>Our methods achieve up to 9% performance gain with 9 times of computational efficiencies.
  arXiv  Detail & Related papers  (2025-08-07T02:51:10Z)
• Stochastic Encodings for Active Feature Acquisition [100.47043816019888]
  Active Feature Acquisition is an instance-wise, sequential decision making problem.<n>The aim is to dynamically select which feature to measure based on current observations, independently for each test instance.<n>Common approaches either use Reinforcement Learning, which experiences training difficulties, or greedily maximize the conditional mutual information of the label and unobserved features, which makes myopic.<n>We introduce a latent variable model, trained in a supervised manner. Acquisitions are made by reasoning about the features across many possible unobserved realizations in a latent space.
  arXiv  Detail & Related papers  (2025-08-03T23:48:46Z)
• 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)
• DEAL: Disentangling Transformer Head Activations for LLM Steering [19.770342907146965]
  We propose a principled causal-attribution framework for identifying behavior-relevant attention heads in transformers.<n>For each head, we train a vector-quantized autoencoder (VQ-AE) on its attention activations.<n>We assess the behavioral relevance of each head by the separability of VQ-AE encodings for behavior-aligned versus behavior-violating responses.
  arXiv  Detail & Related papers  (2025-06-10T02:16:50Z)
• Offline Multi-agent Reinforcement Learning via Score Decomposition [51.23590397383217]
  offline multi-agent reinforcement learning (MARL) faces critical challenges due to distributional shifts and the high dimensionality of joint action spaces.<n>We propose a novel two-stage framework for modeling diverse multi-agent coordination patterns.<n>Our approach provides new insights into offline coordination and equilibrium selection in cooperative multi-agent systems.
  arXiv  Detail & Related papers  (2025-05-09T11:42:31Z)
• 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)
• 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)
• Probe-Free Low-Rank Activation Intervention [26.502232859901167]
  Inference-time interventions that edit the hidden activations have shown promising results in steering the LMs towards desirable generations.<n>This paper proposes a probe-free intervention method FLORAIN for all attention heads in a specific activation layer.
  arXiv  Detail & Related papers  (2025-02-06T13:03:05Z)
• Semantics-Adaptive Activation Intervention for LLMs via Dynamic Steering Vectors [8.761404991620285]
  Activation intervention has emerged as an effective and economical method to modify the behavior of large language models (LLMs)<n>We propose Semantics-Adaptive Dynamic Intervention (SADI), a novel method that constructs a dynamic steering vector to intervene model activations at inference time.<n> Experimental results show that SADI outperforms established baselines by substantial margins, improving task performance without training.
  arXiv  Detail & Related papers  (2024-10-16T06:58:49Z)
• Dynamic Correlation Learning and Regularization for Multi-Label Confidence Calibration [60.95748658638956]
  This paper introduces the Multi-Label Confidence task, aiming to provide well-calibrated confidence scores in multi-label scenarios. Existing single-label calibration methods fail to account for category correlations, which are crucial for addressing semantic confusion. We propose the Dynamic Correlation Learning and Regularization algorithm, which leverages multi-grained semantic correlations to better model semantic confusion.
  arXiv  Detail & Related papers  (2024-07-09T13:26:21Z)
• Steering Llama 2 via Contrastive Activation Addition [41.54815073311959]
  Contrastive Activation Addition (CAA) is a method for steering language models by modifying their activations during forward passes. CAA accurately steers model outputs and sheds light on how high-level concepts are represented in Large Language Models (LLMs)
  arXiv  Detail & Related papers  (2023-12-09T04:40:46Z)
• MCDAL: Maximum Classifier Discrepancy for Active Learning [74.73133545019877]
  Recent state-of-the-art active learning methods have mostly leveraged Generative Adversarial Networks (GAN) for sample acquisition. We propose in this paper a novel active learning framework that we call Maximum Discrepancy for Active Learning (MCDAL) In particular, we utilize two auxiliary classification layers that learn tighter decision boundaries by maximizing the discrepancies among them.
  arXiv  Detail & Related papers  (2021-07-23T06:57:08Z)
• Adaptive Correlated Monte Carlo for Contextual Categorical Sequence Generation [77.7420231319632]
  We adapt contextual generation of categorical sequences to a policy gradient estimator, which evaluates a set of correlated Monte Carlo (MC) rollouts for variance control. We also demonstrate the use of correlated MC rollouts for binary-tree softmax models, which reduce the high generation cost in large vocabulary scenarios.
  arXiv  Detail & Related papers  (2019-12-31T03:01:55Z)

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.