Disentangling Direction and Magnitude in Transformer Representations: A Double Dissociation Through L2-Matched Perturbation Analysis

• URL: http://arxiv.org/abs/2602.11169v1
• Date: Mon, 19 Jan 2026 06:45:04 GMT
• Title: Disentangling Direction and Magnitude in Transformer Representations: A Double Dissociation Through L2-Matched Perturbation Analysis
• Abstract summary: We study a striking cross-over dissociation: angular perturbations cause up to 42.9 more damage to language modeling loss.<n>We find that direction preferentially affects attentional routing, while magnitude modulates processing intensity for fine-grained syntactic judgments.<n>Our results refine the linear representation hypothesis and have implications for model editing and interpretability research.
• Score: 0.42970700836450487
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Transformer hidden states encode information as high-dimensional vectors, yet whether direction (orientation in representational space) and magnitude (vector norm) serve distinct functional roles remains unclear. Studying Pythia-family models, we discover a striking cross-over dissociation: angular perturbations cause up to 42.9 more damage to language modeling loss, while magnitude perturbations cause disproportionately more damage to syntactic processing (20.4% vs.1.6% accuracy drop on subject-verb agreement).This finding is enabled by L2-matched perturbation analysis, a methodology ensuring that an gular and magnitude perturbations achieve identical Euclidean displacements. Causal intervention reveals that angular damage flows substantially through the attention pathways (28.4% loss recovery via attention repair), while magnitude damage flows partly through the LayerNorm pathways(29.9% recovery via LayerNorm repair). These patterns replicate across scales within the Pythia architecture family. These findings provide evidence that direction and magnitude support partially distinct computational roles in LayerNorm based architectures. The direction preferentially affects attentional routing, while magnitude modulates processing intensity for fine-grained syntactic judgments. We find different patterns in RMSNorm-based architectures, suggesting that the dissociation depends on architectural choices. Our results refine the linear representation hypothesis and have implications for model editing and interpretability research

Related papers

• Orientation, not magnitude: the causal structure of task-vector interference in merged language models [0.0]
  We show that task arithmetic works until it doesn't, and the field diagnoses why with magnitudes.<n>An exact decomposition of the layerwise flux shows it is dominated by amplifying transport of the existing cross-term.<n>The striking +-15% "universality" of naive bfloat16 generation turns out to be quantization roughness.
  arXiv  Detail & Related papers  (2026-08-12T08:40:24Z)
• Actionable Activation Directions for Detecting and Mitigating Emergent Misalignment Across Language Model Families [0.0]
  Fine-tuning language models on insecure code induces emergent misalignment with poorly understood internal structure.<n>We investigate whether this misalignment corresponds to a causally actionable activation-space direction shared across architectures.
  arXiv  Detail & Related papers  (2026-06-18T13:39:59Z)
• Structure-Aware Distributed Backdoor Attacks in Federated Learning [2.04367431902848]
  This paper analyzes the coupling relationship between model architectures and backdoor perturbations.<n> Experimental results show that model architecture significantly influences the propagation and aggregation of perturbations.
  arXiv  Detail & Related papers  (2026-03-04T09:19:32Z)
• TopoCurate:Modeling Interaction Topology for Tool-Use Agent Training [53.93696896939915]
  Training tool-use agents typically rely on Supervised Fine-Tuning (SFT) on successful trajectories and Reinforcement Learning (RL) on pass-rate-selected tasks.<n>We propose TopoCurate, an interaction-aware framework that projects multi-trial rollouts from the same task into a unified semantic quotient topology.<n>TopoCurate achieves consistent gains of 4.2% (SFT) and 6.9% (RL) over state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-03-02T10:38:54Z)
• MirrorLA: Reflecting Feature Map for Vision Linear Attention [49.41670925034762]
  Linear attention significantly reduces the computational complexity of Transformers from quadratic to linear, yet it consistently lags behind softmax-based attention in performance.<n>We propose MirrorLA, a geometric framework that substitutes passive truncation with active reorientation.<n>MirrorLA achieves state-of-the-art performance across standard benchmarks, demonstrating that strictly linear efficiency can be achieved without compromising representational fidelity.
  arXiv  Detail & Related papers  (2026-02-04T09:14:09Z)
• Simulated Adoption: Decoupling Magnitude and Direction in LLM In-Context Conflict Resolution [3.0242762196828448]
  Large Language Models (LLMs) frequently prioritize conflicting in-context information over pre-existing parametric memory.<n>We show that models do not "unlearn" or suppress the magnitude of internal truths but rather employ a mechanism of geometric displacement.
  arXiv  Detail & Related papers  (2026-02-04T06:13:11Z)
• Causal Discovery with Mixed Latent Confounding via Precision Decomposition [0.0]
  Differentiable and score-based DAG learners can misinterpret global latent effects as causal edges, while latent-variable graphical models recover only undirected structure.<n>We propose textscDCL-DECOR, a modular, precision-led pipeline that separates these roles.
  arXiv  Detail & Related papers  (2025-12-31T08:03:41Z)
• Robust Self-Supervised LiDAR Odometry via Representative Structure Discovery and 3D Inherent Error Modeling [67.75095378830694]
  We develop a two-stage odometry estimation network, where we obtain the ego-motion by estimating a set of sub-region transformations. In this paper, we aim to alleviate the influence of unreliable structures in training, inference and mapping phases. Our two-frame odometry outperforms the previous state of the arts by 16%/12% in terms of translational/rotational errors.
  arXiv  Detail & Related papers  (2022-02-27T12:52:27Z)
• Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence [100.6913091147422]
  Existing rotated object detectors are mostly inherited from the horizontal detection paradigm. In this paper, we are motivated to change the design of rotation regression loss from induction paradigm to deduction methodology.
  arXiv  Detail & Related papers  (2021-06-03T14:29:19Z)
• Understanding and Diagnosing Vulnerability under Adversarial Attacks [62.661498155101654]
  Deep Neural Networks (DNNs) are known to be vulnerable to adversarial attacks. We propose a novel interpretability method, InterpretGAN, to generate explanations for features used for classification in latent variables. We also design the first diagnostic method to quantify the vulnerability contributed by each layer.
  arXiv  Detail & Related papers  (2020-07-17T01:56:28Z)
• Sparse representation for damage identification of structural systems [11.397437423613418]
  We propose a novel two-stage sensitivity analysis-based framework for both model updating and sparse damage identification. A sparse representation pipeline built on a quasi-$ell$ method is then presented for damage and localization quantification. Results show that the proposed approach is capable of both localizing and quantifying structural damage with high accuracy.
  arXiv  Detail & Related papers  (2020-06-06T18:04:35Z)
• A Critical View of the Structural Causal Model [89.43277111586258]
  We show that one can identify the cause and the effect without considering their interaction at all. We propose a new adversarial training method that mimics the disentangled structure of the causal model. Our multidimensional method outperforms the literature methods on both synthetic and real world datasets.
  arXiv  Detail & Related papers  (2020-02-23T22:52:28Z)

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.