Curveball Steering: The Right Direction To Steer Isn't Always Linear

• URL: http://arxiv.org/abs/2603.09313v2
• Date: Wed, 11 Mar 2026 02:43:46 GMT
• Title: Curveball Steering: The Right Direction To Steer Isn't Always Linear
• Abstract summary: 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.
• Score: 7.264147948470562
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Activation steering is a widely used approach for controlling large language model (LLM) behavior by intervening on internal representations. Existing methods largely rely on the Linear Representation Hypothesis, assuming behavioral attributes can be manipulated using global linear directions. In practice, however, such linear interventions often behave inconsistently. We question this assumption by analyzing the intrinsic geometry of LLM activation spaces. Measuring geometric distortion via the ratio of geodesic to Euclidean distances, we observe substantial and concept-dependent distortions, indicating that activation spaces are not well-approximated by a globally linear geometry. Motivated by this, we propose "Curveball steering", a nonlinear steering method based on polynomial kernel PCA that performs interventions in a feature space, better respecting the learned activation geometry. Curveball steering consistently outperforms linear PCA-based steering, particularly in regimes exhibiting strong geometric distortion, suggesting that geometry-aware, nonlinear steering provides a principled alternative to global, linear interventions.

Related papers

• Beyond Linear Activation Steering: Invertible Latent Transformations for Controlling LLM Behavior [13.073472989807675]
  INNSteer is a nonlinear activation steering framework based on invertible latent transformations.<n>It consistently improves model control over linear, transport-based, and nonlinear steering baselines.
  arXiv  Detail & Related papers  (2026-06-07T05:01:25Z)
• LAST: Bridging Vision-Language and Action Manifolds via Gromov-Wasserstein Alignment [71.3712075841632]
  LAST reconstructs the action space to establish local metric compatibility with the Vision-Language-Action modality.<n> LAST enables VLA models with superior convergence and generalizability.
  arXiv  Detail & Related papers  (2026-05-27T03:48:10Z)
• General Covariant Action Modeling: Constructing Generalized Manifolds via Spatio-Temporal Decoupling [48.559681359374146]
  Generalized Action Manifold (GAM) framework enforces general covariance through structural disentanglement.<n>GAM enables superior transfer and robustness capabilities, outperforming geometry-agnostic baselines.
  arXiv  Detail & Related papers  (2026-05-27T03:38:15Z)
• Riemannian-Manifold Steering: Geometry-Aware Generative Autoencoders for Label-Free Steering [3.854000879223449]
  We recast manifold steering as textbfRiemannian geodesic computation on activation space.<n>We approximate this with a learned encoder trained on output distances over a small concept-token schema.
  arXiv  Detail & Related papers  (2026-05-24T08:41:56Z)
• A Mean Curvature Approach to Boundary Detection: Geometric Insights for Unsupervised Learning [52.452902154360565]
  We introduce Mean Curvature Boundary Points (MCBP), a novel geometric framework grounded in Geometric Machine Learning.<n>MCBP explicitly models the intrinsic curvature of the data manifold and computes pointwise mean curvature without requiring principled manifold parametrization.<n>Experiments on synthetic and real-world datasets demonstrate that MCBP consistently improves clustering performance.
  arXiv  Detail & Related papers  (2026-05-05T20:19:09Z)
• The Cylindrical Representation Hypothesis for Language Model Steering [57.97381760521523]
  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.
  arXiv  Detail & Related papers  (2026-05-03T12:26:13Z)
• Local Linearity of LLMs Enables Activation Steering via Model-Based Linear Optimal Control [2.752817022620644]
  Inference-time LLM alignment methods offer an alternative to fine-tuning by directly modifying activations during generation.<n>We show that layer-wise dynamics across multiple LLM architectures and scales are well-approximated by locally-linear models.<n>We derive theoretical bounds on setpoint tracking error, enabling formal guarantees on steering performance.
  arXiv  Detail & Related papers  (2026-04-21T03:09:46Z)
• Curvature-Aware PCA with Geodesic Tangent Space Aggregation for Semi-Supervised Learning [52.452902154360565]
  GTSA-PCA is a geometric extension of Principal Component Analysis.<n>It integrates curvature awareness and geodesic consistency within a unified spectral framework.<n>Our results position GTSA-PCA as a principled bridge between statistical and geometric approaches to dimensionality reduction.
  arXiv  Detail & Related papers  (2026-04-20T20:36:36Z)
• Spherical Steering: Geometry-Aware Activation Rotation for Language Models [15.078810641141295]
  Inference-time steering has emerged as a promising paradigm for controlling language models (LMs) without the cost of retraining.<n>In this work, we explore Spherical Steering, a training-free primitive that resolves this trade-off through activation rotation.<n>Our method rotates activations along a geodesic toward a target direction, guiding the activation toward the target concept while preserving the integrity of the signal.
  arXiv  Detail & Related papers  (2026-02-09T00:15:47Z)
• Orthogonalized Policy Optimization:Decoupling Sampling Geometry from Optimization Geometry in RLHF [0.0]
  Large language model alignment objectives are often presented as a collection of distinct algorithms, such as PPO, DPO, IPO, and their variants.<n>In this work, we argue that this diversity obscures a simpler underlying structure.<n>We show that this entanglement is not merely a modeling convenience but a source of systematic instability.
  arXiv  Detail & Related papers  (2026-01-18T13:57:44Z)
• Quantized nonlinear kink movement through topological boundary state instabilities [0.0]
  We study a nonlinear dimer-chain model that exhibits quantized transport of kinks under a periodic modulation of a pumping parameter.<n>We show that instabilities in these boundary modes are the driving mechanism behind the observed kink motion.<n>Our results unify concepts from linear topology and nonlinear dynamics to establish a framework for quantized transport in nonlinear media.
  arXiv  Detail & Related papers  (2025-08-16T16:46:19Z)
• Enforcing Latent Euclidean Geometry in Single-Cell VAEs for Manifold Interpolation [79.27003481818413]
  We introduce FlatVI, a training framework that regularises the latent manifold of discrete-likelihood variational autoencoders towards Euclidean geometry.<n>By encouraging straight lines in the latent space to approximate geodesics on the decoded single-cell manifold, FlatVI enhances compatibility with downstream approaches.
  arXiv  Detail & Related papers  (2025-07-15T23:08:14Z)
• Convergence of TD(0) under Polynomial Mixing with Nonlinear Function Approximation [49.1574468325115]
  Temporal Difference Learning (TD(0)) is fundamental in reinforcement learning.<n>We provide the first high-probability, finite-sample analysis of vanilla TD(0) on mixing Markov data.
  arXiv  Detail & Related papers  (2025-02-08T22:01:02Z)
• RLE: A Unified Perspective of Data Augmentation for Cross-Spectral Re-identification [59.5042031913258]
  Non-linear modality discrepancy mainly comes from diverse linear transformations acting on the surface of different materials. We propose a Random Linear Enhancement (RLE) strategy which includes Moderate Random Linear Enhancement (MRLE) and Radical Random Linear Enhancement (RRLE) The experimental results not only demonstrate the superiority and effectiveness of RLE but also confirm its great potential as a general-purpose data augmentation for cross-spectral re-identification.
  arXiv  Detail & Related papers  (2024-11-02T12:13:37Z)
• KEEC: Koopman Embedded Equivariant Control [29.738391644702947]
  An efficient way to control systems with unknown nonlinear dynamics is to find an appropriate embedding or representation.<n>Koopman Embedded Equivariant Control (KEEC) learns an embedding of the states and vector fields such that a Koopman operator is approximated as the latent dynamics.<n>Our algorithm achieves superior performances in the experiments conducted on various control domains.
  arXiv  Detail & Related papers  (2023-12-04T00:11:27Z)
• Stochastic Nonlinear Control via Finite-dimensional Spectral Dynamic Embedding [20.43835169613882]
  This paper proposes an approach, Spectral Dynamics Embedding Control (SDEC), to optimal control for nonlinear systems.<n>It reveals an infinite-dimensional feature representation induced by the system's nonlinear dynamics, enabling a linear representation of the state-action value function.<n>For practical implementation, this representation is approximated using finite-dimensional truncations.
  arXiv  Detail & Related papers  (2023-04-08T04:23:46Z)
• Policy Gradient for Reinforcement Learning with General Utilities [50.65940899590487]
  In Reinforcement Learning (RL), the goal of agents is to discover an optimal policy that maximizes the expected cumulative rewards. Many supervised and unsupervised RL problems are not covered in the Linear RL framework. We derive the policy gradient theorem for RL with general utilities.
  arXiv  Detail & Related papers  (2022-10-03T14:57:46Z)
• A dynamical systems based framework for dimension reduction [0.0]
  We propose a framework for learning a low-dimensional representation of data based on nonlinear dynamical systems. In the DDR model, each point is evolved via a nonlinear flow towards a lower-dimensional subspace. We show how the DDR method can be trained using a gradient-based optimization method.
  arXiv  Detail & Related papers  (2022-04-18T04:02:11Z)
• Provably Correct Optimization and Exploration with Non-linear Policies [65.60853260886516]
  ENIAC is an actor-critic method that allows non-linear function approximation in the critic. We show that under certain assumptions, the learner finds a near-optimal policy in $O(poly(d))$ exploration rounds. We empirically evaluate this adaptation and show that it outperforms priors inspired by linear methods.
  arXiv  Detail & Related papers  (2021-03-22T03:16:33Z)

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.