Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling

• URL: http://arxiv.org/abs/2608.03842v1
• Date: Tue, 04 Aug 2026 15:48:55 GMT
• Title: Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling
• Abstract summary: We show that when a language model fails on surface-perturbed input, "which layer is responsible" has three natural operationalizations.<n>We identify two propagation regimes - spike-and-suppress and late-accumulation.<n>A fixed-harness layer sweep across four models (3.8-8B) confirms the core prediction on chain-of-thought GSM8K.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: When a language model fails on surface-perturbed input (typos, OCR noise, homophones), "which layer is responsible" has three natural operationalizations: where representations diverge most (sensitivity), where restoring clean activations recovers the prediction (causality), and where a small adapter can repair the damage (compensatory capacity) - and we show these three layer maps dissociate. Across a five-model panel we identify two propagation regimes - spike-and-suppress (Phi-3.5, Gemma-2-9B) and late-accumulation (Llama-3, Mistral, Qwen2.5-7B) - and on the two models meeting an 80% identity-patch gate, sensitivity and causality are anti-correlated (rho = -0.72 to -0.88). Within-family scaling on Qwen2.5 (1.5B to 14B) shows the late-accumulation signature strengthening monotonically with scale, corroborated on a second family. We propose cascade disruption as the mechanism behind the dissociation: adapters placed at causally implicated early layers break intact downstream computation, making diagnostic-flagged sites the worst adapter placements. A fixed-harness layer sweep across four models (3.8-8B) confirms the core prediction on chain-of-thought GSM8K - the flagged sites are the most damaging adapter windows on every adjudicable model - and is sign-consistent but strongly attenuated on a multiple-choice control, consistent with damage that compounds with generation length. The sweep yields practical guidance: a training-free LRD pre-screen and a default-deepest placement rule, though absolute gains over no-adapter baselines remain small. Finally, apparent gains from a representation-stability loss reverse under an adequate generation budget - truncated chain-of-thought had been scored as empty - a methodological warning for any intervention evaluated on chain-of-thought tasks.

Related papers

• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• SafeCA: Safe Cross-Attention Localization and Regulation for Text-to-Video Jailbreak Defense [77.70598852234149]
  Text-to-Video (T2V) generative models are vulnerable to jailbreak attacks in real-world deployment.<n>We propose SafeCA, a feature-level defense mechanism for safe cross-attention localization and regularization.<n> Experimental results show that SafeCA reduces the jailbreak success rate by about 20% on mainstream T2V models.
  arXiv  Detail & Related papers  (2026-08-11T14:01:24Z)
• Wrong Before Right: Late Rescue and Interface Failure in Aligned Language Models [2.1291393112295935]
  We study how correctness is assembled inside aligned language models.<n>Using layer-wise difference-in-differences (DiD) trajectories over polarity-controlled minimal pairs, we identify the wrong-dip.<n>We verify this causally with patchscope-style activation transplantation across 17 models, three families, and 64x scale (0.5B-32B)
  arXiv  Detail & Related papers  (2026-07-06T03:51:24Z)
• DriftScope: Measuring The Hidden Effects of Diffusion Model Adaptation [50.584834483403675]
  Adapting pre-trained text-to-image diffusion models, whether to learn new visual concepts or erase unwanted ones, is routinely evaluated on its intended effects alone.<n>We demonstrate that adaptation systematically damages semantically unrelated concepts in ways that aggregate metrics structurally cannot surface.<n>We introduce DriftScope, a prompt-level diagnostic tool that takes any two model checkpoints and returns a ranked list of tokens whose visual concepts have shifted most between them.
  arXiv  Detail & Related papers  (2026-06-30T20:57:51Z)
• DART: A Vision-Language Foundation Model for Comprehensive Rope Condition Monitoring [2.8993790400286876]
  We present DART (Damage Assessment via Rope Transformer), a vision-supervised foundation model that addresses the full rope inspection through a unified multi-task architecture.<n>Three architectural innovations drive the model's versatility: HD-MASK, a saliency-guided masking strategy that focuses self-language reconstruction on damage-dense patches.<n>DART functions as a general-purpose CM backbone that goes well beyond classification, providing actionable inspection intelligence from a single shared representation.
  arXiv  Detail & Related papers  (2026-05-06T14:12:02Z)
• Fragile Reasoning: A Mechanistic Analysis of LLM Sensitivity to Meaning-Preserving Perturbations [2.7946918847372277]
  Large language models demonstrate strong performance on mathematical reasoning benchmarks, yet remain surprisingly fragile to meaning-preserving surface perturbations.<n>We systematically evaluate three open-weight LLMs, Mistral-7B, Llama-3-8B, and Qwen2.5-7B, on 677 GSM8K problems paired with semantically equivalent variants.<n>All three models exhibit substantial answer-flip rates (28.8%-45.1%), with number paraphrasing consistently more disruptive than name swaps.
  arXiv  Detail & Related papers  (2026-04-02T05:30:20Z)
• Engineering Verifiable Modularity in Transformers via Per-Layer Supervision [0.0]
  Ablating an attention head identified as critical for capitalization produces minimal behavioral change because distributed redundancy compensates for damage.<n>We demonstrate that architectural interventions can expose hidden modularity.<n>This es tablishes a methodology for transforming interpretability from passive observation to active control.
  arXiv  Detail & Related papers  (2026-03-08T05:18:14Z)
• BURN: Backdoor Unlearning via Adversarial Boundary Analysis [73.14147934175604]
  Backdoor unlearning aims to remove backdoor-related information while preserving the model's original functionality.<n>We propose Backdoor Unlearning via adversaRial bouNdary analysis (BURN), a novel defense framework that integrates false correlation decoupling, progressive data refinement, and model purification.
  arXiv  Detail & Related papers  (2025-07-14T17:13:06Z)
• Backdoor Cleaning without External Guidance in MLLM Fine-tuning [76.82121084745785]
  Believe Your Eyes (BYE) is a data filtering framework that leverages attention entropy patterns as self-supervised signals to identify and filter backdoor samples.<n>It achieves near-zero attack success rates while maintaining clean-task performance.
  arXiv  Detail & Related papers  (2025-05-22T17:11:58Z)
• Overlap-Aware Feature Learning for Robust Unsupervised Domain Adaptation for 3D Semantic Segmentation [9.578322021478426]
  3D point cloud semantic segmentation (PCSS) is a cornerstone for environmental perception in robotic systems and autonomous driving.<n>Existing methods critically overlook the inherent vulnerability to real-world perturbations (e.g., snow, fog, rain) and adversarial distortions.<n>This work first identifies two intrinsic limitations that undermine current PCSS-UDA robustness.<n>We propose a tripartite framework consisting of: 1) a robustness evaluation model quantifying resilience against adversarial attack/corruption types through robustness metrics; 2) an invertible attention alignment module (IAAM) enabling bidirectional domain mapping while preserving discriminative structure via attention-guided overlap suppression; and
  arXiv  Detail & Related papers  (2025-04-02T12:16:23Z)
• Uncertainty-Aware Adaptation for Self-Supervised 3D Human Pose Estimation [70.32536356351706]
  We introduce MRP-Net that constitutes a common deep network backbone with two output heads subscribing to two diverse configurations. We derive suitable measures to quantify prediction uncertainty at both pose and joint level. We present a comprehensive evaluation of the proposed approach and demonstrate state-of-the-art performance on benchmark datasets.
  arXiv  Detail & Related papers  (2022-03-29T07:14:58Z)
• The KFIoU Loss for Rotated Object Detection [115.334070064346]
  In this paper, we argue that one effective alternative is to devise an approximate loss who can achieve trend-level alignment with SkewIoU loss. Specifically, we model the objects as Gaussian distribution and adopt Kalman filter to inherently mimic the mechanism of SkewIoU. The resulting new loss called KFIoU is easier to implement and works better compared with exact SkewIoU.
  arXiv  Detail & Related papers  (2022-01-29T10:54:57Z)

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.