The Cylindrical Representation Hypothesis for Language Model Steering

• URL: http://arxiv.org/abs/2605.01844v1
• Date: Sun, 03 May 2026 12:26:13 GMT
• Title: The Cylindrical Representation Hypothesis for Language Model Steering
• Abstract summary: We show that a central axis captures the main difference between concept absence and presence and drives concept generation.<n>We formalize this as the Cylindrical Representation Hypothesis (CRH)<n>Our experiments verify the existence of the cylindrical structure and demonstrate that CRH provides a valid and practical way to interpret model steering behavior in real settings.
• Score: 57.97381760521523
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Steering is a widely used technique for controlling large language models, yet its effects are often unstable and hard to predict. Existing theoretical accounts are largely based on the Linear Representation Hypothesis (LRH). While LRH assumes that concepts can be orthogonalized for lossless control, this idealized mapping fails in real representations and cannot account for the observed unpredictability of steering. By relaxing LRH's orthogonality assumption while preserving linear representations, we show that overlapping concept contributions naturally yield a sample-specific axis-orthogonal structure. We formalize this as the Cylindrical Representation Hypothesis (CRH). In CRH, a central axis captures the main difference between concept absence and presence and drives concept generation. A surrounding normal plane controls steering sensitivity by determining how easily the axis can activate the target concept. Within this plane, only specific sensitive sectors strongly facilitate concept activation, while other sectors can suppress or delay it. While the surrounding normal plane can be reliably identified from difference vectors, the sensitive sector cannot, introducing intrinsic uncertainty at the sector level. This uncertainty provides a principled explanation for why steering outcomes often fluctuate even when using well-aligned directions. Our experiments verify the existence of the cylindrical structure and demonstrate that CRH provides a valid and practical way to interpret model steering behavior in real settings: https://github.com/mbzuai-nlp/CRH.

Related papers

• Non-normal spectral signatures of instability in neural network training dynamics [0.5307735857768683]
  We show that the linearized update operators for practically useds are generically non-normal.<n>Applying non-normal stability theory to these operators, we derive a conservative pseudospectral precursor bound in which (V) serves as an early-warning indicator of transient amplification.
  arXiv  Detail & Related papers  (2026-05-22T10:36:48Z)
• Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates [6.396157607535812]
  Graph-based surrogate models provide fast alternatives to high-fidelity CFD solvers, but their opaque latent spaces and limited controllability restrict use in safety-critical settings.<n>We ask whether phase drift can instead be corrected post hoc by manipulating the latent space of a frozen surrogate.<n>We propose a phase-steering framework for pretrained graph-based CFD models that combines the right representation with the right intervention mechanism.
  arXiv  Detail & Related papers  (2026-03-28T07:03:44Z)
• Curveball Steering: The Right Direction To Steer Isn't Always Linear [7.264147948470562]
  Activation steering is a widely used approach for controlling large language model (LLM) behavior by intervening on internal representations.<n>We propose "Curveball steering", a nonlinear steering method based on PCA kernel that performs interventions in a feature space.
  arXiv  Detail & Related papers  (2026-03-10T07:45:35Z)
• Concept Heterogeneity-aware Representation Steering [6.247724933679344]
  Representation steering offers a mechanism for controlling the behavior of large language models (LLMs) by intervening on internal activations at inference time.<n>Most existing methods rely on a single global steering direction, typically obtained via difference-in-means over contrastive datasets.<n>In this work, we view representation steering through the lens of optimal transport (OT), noting that standard difference-in-means steering implicitly corresponds to the OT map between two unimodal Gaussian distributions.
  arXiv  Detail & Related papers  (2026-02-13T11:07:23Z)
• On the Identifiability of Steering Vectors in Large Language Models [0.0]
  Activation steering methods are widely used to control large language model behavior.<n>This interpretation implicitly assumes steering directions are identifiable and uniquely recoverable from input-output behavior.<n>We prove that steering vectors are fundamentally non-identifiable due to large equivalence classes of behaviorally indistinguishable interventions.
  arXiv  Detail & Related papers  (2026-02-06T15:53:50Z)
• Towards Inference-time Scaling for Continuous Space Reasoning [55.40260529506702]
  Inference-time scaling has proven effective for text-based reasoning in large language models.<n>This paper investigates whether such established techniques can be successfully adapted to reasoning in the continuous space.<n>We demonstrate the feasibility of generating diverse reasoning paths through dropout-based sampling.
  arXiv  Detail & Related papers  (2025-10-14T05:53:41Z)
• The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence [57.57786477441956]
  Prior work suggests that a single refusal direction in the model's activation space determines whether an LLM refuses a request.<n>We propose a novel gradient-based approach to representation engineering and use it to identify refusal directions.<n>We show that refusal mechanisms in LLMs are governed by complex spatial structures and identify functionally independent directions.
  arXiv  Detail & Related papers  (2025-02-24T18:52:59Z)
• Log-linear Guardedness and its Implications [116.87322784046926]
  Methods for erasing human-interpretable concepts from neural representations that assume linearity have been found to be tractable and useful. This work formally defines the notion of log-linear guardedness as the inability of an adversary to predict the concept directly from the representation. We show that, in the binary case, under certain assumptions, a downstream log-linear model cannot recover the erased concept.
  arXiv  Detail & Related papers  (2022-10-18T17:30:02Z)
• Robustness and Accuracy Could Be Reconcilable by (Proper) Definition [109.62614226793833]
  The trade-off between robustness and accuracy has been widely studied in the adversarial literature. We find that it may stem from the improperly defined robust error, which imposes an inductive bias of local invariance. By definition, SCORE facilitates the reconciliation between robustness and accuracy, while still handling the worst-case uncertainty.
  arXiv  Detail & Related papers  (2022-02-21T10:36:09Z)
• Discovering Latent Causal Variables via Mechanism Sparsity: A New Principle for Nonlinear ICA [81.4991350761909]
  Independent component analysis (ICA) refers to an ensemble of methods which formalize this goal and provide estimation procedure for practical application. We show that the latent variables can be recovered up to a permutation if one regularizes the latent mechanisms to be sparse.
  arXiv  Detail & Related papers  (2021-07-21T14:22:14Z)

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.