Decomposition of Evidence, Contradiction, and Fragility in Perturbation Responses

• URL: http://arxiv.org/abs/2608.12935v1
• Date: Thu, 13 Aug 2026 08:13:22 GMT
• Title: Decomposition of Evidence, Contradiction, and Fragility in Perturbation Responses
• Abstract summary: We introduce DECAF, which routes aligned, opposed, and endpoint-null responses into evidence E, contradiction C, and F.<n>The decomposition preserves ordinary magnitude exactly, Abs = E + C + F, and is unique under endpoint-relative axioms.<n>Changing only the reveal path increases total response by nearly 80%, yet evidence barely changes while fragility grows by more than 4x.
• Score: 1.1162481475388237
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Perturbation methods explain model decisions by measuring prediction changes under altered inputs, but response magnitude tells us only how much a model reacts, not what that reaction means. The same magnitude can support the final factual-counterfactual difference, oppose it, or arise strongly along the perturbation path yet vanish at the endpoint. We therefore track how the contrast develops as paired inputs are progressively revealed, using the final contrast to interpret the trajectory. We introduce DECAF (Decomposition of Evidence, Contradiction, And Fragility), which routes aligned, opposed, and endpoint-null responses into evidence E, contradiction C, and fragility F. The decomposition preserves ordinary magnitude exactly, Abs = E + C + F, and is unique under endpoint-relative axioms. Across controlled vision and tabular settings, the three components track independently measured behavior. In a 72-model ImageNet-9 audit, we compare cases with nearly identical response magnitude but different independently measured behaviors. The largest DECAF component agrees with an observed behavior in 96.4% of cases, compared with 35.0% for magnitude alone. Changing only the reveal path increases total response by nearly 80%, yet evidence barely changes while fragility grows by more than 4x. On FunnyBirds and ImageNet-1k, short forward-only DECAF trajectories outperform the tested general-purpose attribution baselines. On a 1B-scale DINOv2 model, a short trajectory matches a strong gradient-based baseline with 4.75x lower wall time and 2.36x lower peak memory.

Related papers

• A Calibration Audit of Confidence in Feed-Forward 3D Reconstruction [45.496856835431885]
  3D reconstruction models emit per-pixel confidence that downstream systems read as a reliability signal.<n>We score the confidence on four properties, how well it ranks error, whether its level is right on average, whether it holds across the confidence range, and whether its intervals cover the truth.<n>The confidence ranks error well, but the predicted uncertainty is too low when it is read under conditions that are not exactly those of training.
  arXiv  Detail & Related papers  (2026-08-30T10:21:34Z)
• The Imperfective Paradox Is Not Necessarily in Large Language Models: A Benchmark Failure Before a Model Failure [42.25959731143407]
  We examine the benchmark and conclusions and show that it is substantially affected by conceptual and evaluation mis-specifications.<n>Under a strict NLI standard, 76% of Group A instances do not explicitly rule out culmination.<n>We formulate event-semantic NLI as a Multi-step Reasoning Problem and assess both intermediate semantic decisions and final predictions.
  arXiv  Detail & Related papers  (2026-08-25T18:00:18Z)
• 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)
• Visual Credit Audit for Multimodal Spatial Reasoning [70.16915309526443]
  Visual Credit Audit separates two Estimands: whether the benchmark image gives the model's declared decision more support than text-only and blank controls, and whether the model responds to relation-specific visual evidence.<n>Applying labels yields dependence-credited correctness (D-CC)<n>Across four open MLLMs and two spatial benchmarks, 12.73-26.25% of decisions are correct yet uncredited.
  arXiv  Detail & Related papers  (2026-07-29T15:55:31Z)
• Robust for the Wrong Reasons: The Representational Geometry of LLM Robustness to Science Skepticism [3.993449663756884]
  Large language models (LLMs) are increasingly consulted on contested scientific questions.<n>We show that models show three distinct policies under the same skeptical pressure.<n>We synthesize these into a four-way taxonomy separating active from accidental robustness.
  arXiv  Detail & Related papers  (2026-07-02T09:40:52Z)
• Measuring and Improving Behavioral Consistency in Large Language Models through Fact-Heuristic-Emotion State Enforcement [4.226475360842309]
  Large language models (LLMs) can give different answers to the same decision problem across runs, and reverse a decision when their own prior answer returns as context.<n>We ask whether this instability can be measured and partially reduced without changing model weights.<n>We test the Cognitive Kernel Model (CKM), a prompt-level state-enforcement layer.
  arXiv  Detail & Related papers  (2026-06-05T12:36:36Z)
• Disagreement-Based Cross-Model Routing for Implicit Video Question Answering [0.0]
  We study multiple-choice video question answering on the ImplicitQA benchmark.<n>On this benchmark a single frontier video LLM already operates near its accuracy ceiling.<n>We propose disagreement-based cross-model routing, a pure inference-time procedure that requires no labels and no training.
  arXiv  Detail & Related papers  (2026-05-31T05:56:27Z)
• One Token Away from Collapse: The Fragility of Instruction-Tuned Helpfulness [12.183451602438753]
  We show that simple lexical constraints (banning a single punctuation character or common word) cause instruction-tuned LLMs to collapse their responses.<n>We show that base models show no systematic collapse under identical constraints, with effects that are small, noisy, and bidirectional.
  arXiv  Detail & Related papers  (2026-04-14T17:40:01Z)
• How Independent are Large Language Models? A Statistical Framework for Auditing Behavioral Entanglement and Reweighting Verifier Ensembles [46.63622714488747]
  Shared pretraining data, distillation, and alignment pipelines can induce hidden behavioral dependencies, latent entanglement.<n>In practice, this manifests as correlated reasoning patterns and synchronized failures.<n>We develop a statistical framework for auditing behavioral entanglement among black-box language models.
  arXiv  Detail & Related papers  (2026-04-08T23:32:06Z)
• Causal Understanding by LLMs: The Role of Uncertainty [43.87879175532034]
  Recent papers show LLMs achieve near-random accuracy in causal relation classification.<n>We investigate whether pretraining exposure to causal examples improves causal understanding.
  arXiv  Detail & Related papers  (2025-09-24T13:06:35Z)
• What's the Harm? Sharp Bounds on the Fraction Negatively Affected by Treatment [58.442274475425144]
  We develop a robust inference algorithm that is efficient almost regardless of how and how fast these functions are learned. We demonstrate our method in simulation studies and in a case study of career counseling for the unemployed.
  arXiv  Detail & Related papers  (2022-05-20T17:36:33Z)
• An Analysis of the Adaptation Speed of Causal Models [80.77896315374747]
  Recently, Bengio et al. conjectured that among all candidate models, $G$ is the fastest to adapt from one dataset to another. We investigate the adaptation speed of cause-effect SCMs using convergence rates from optimization. Surprisingly, we find situations where the anticausal model is advantaged, falsifying the initial hypothesis.
  arXiv  Detail & Related papers  (2020-05-18T23:48:56Z)

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.