arXiv subjects

Yongming Lu

Publications and source records attributed to Yongming Lu.

3 recordsLinked to original sources

arXiv2026

A 3D VTI factored eikonal solver using six-tetrahedron pyramidal stencil

Accurate traveltime computation for the eikonal equation is essential in seismic applications such as tomography and migration. The fast sweeping method (FSM) is widely used because of its unconditional stability and computational efficiency. We develop a 3D fast sweeping solver for vertical transverse isotropic (VTI) media that combines multiplicative factorization with a six-tetrahedron pyramidal stencil. The factorization removes the point-source singularity, while the stencil improves local accuracy. In the unfactored formulation, the six-tetrahedron finite-difference scheme requires solving only quadratic equations. After factorization, however, the local update for the perturbation factor becomes quartic, and the update systems on oblique stencil faces are substantially more complicated than those on non-oblique faces. To resolve these difficulties, we solve the update equation using Ferrari's method with a robust root-selection strategy and derive complete update formulas for all oblique-face configurations. For horizontally constrained faces, the characteristic constraint reduces to a linear relation by exploiting the structure of the VTI Hamiltonian, so the local system still reduces to a quartic equation. For mixed horizontal-vertical constrained faces, we design a bisection-based iterative solver. Numerical examples show that the proposed method effectively suppresses source-related errors and improves traveltime accuracy.

physics.geo-ph↗

arXiv2026

Auditing Stealth Sycophancy in Mental-Health Dialogue: Structured Clinical-State Diagnostics and Clean Matched Benchmarks

Mental-health dialogue models are increasingly evaluated by AI-based evaluators, yet these evaluators often treat surface empathy, supportiveness, or fluency as evidence of safety. In this paper, we study a hidden failure mode that we call implicit sycophancy: a response may appear empathetic while implicitly reinforcing catastrophizing, avoidance, hopeless prediction, or CBT-style labeling. To examine this problem, we introduce a diagnostic benchmark for implicit-sycophancy detection, built from three representative mental-health dialogue sources covering everyday peer support, counseling-style emotional support, and crisis-oriented interaction, and further construct a leakage-audited clean single-response matched benchmark with 500 contexts and 1,500 matched response windows. We then propose Dynamic Emotional Signature Graphs (DESG), a structured offline audit framework that separates LLM-based state extraction from final scoring and evaluates clinical direction through semantic, affective, and cognitive-distortion state transitions rather than free-form LLM judgment. Unlike metadata, surface-style, lexical, embedding, and rubric-LLM baselines, DESG scores the direction of clinical-state change induced by a response; on the leakage-audited clean matched benchmark, DESG-StateRisk improves over the strongest non-DESG baseline by 0.0488 macro-F1 and achieves the best harmful-risk detection result. These results suggest that evaluating implicit sycophancy requires explicit clinical-state modeling together with leakage checks, shortcut controls, and competitive baselines.

cs.CL↗

arXiv2025

TECP: Token-Entropy Conformal Prediction for LLMs

Uncertainty quantification (UQ) for open-ended language generation remains a critical yet underexplored challenge, especially under black-box constraints where internal model signals are inaccessible. In this paper, we introduce Token-Entropy Conformal Prediction (TECP), a novel framework that leverages token-level entropy as a logit-free, reference-free uncertainty measure and integrates it into a split conformal prediction (CP) pipeline to construct prediction sets with formal coverage guarantees. Unlike existing approaches that rely on semantic consistency heuristics or white-box features, TECP directly estimates epistemic uncertainty from the token entropy structure of sampled generations and calibrates uncertainty thresholds via CP quantiles to ensure provable error control. Empirical evaluations across six large language models and two benchmarks (CoQA and TriviaQA) demonstrate that TECP consistently achieves reliable coverage and compact prediction sets, outperforming prior self-consistency-based UQ methods. Our method provides a principled and efficient solution for trustworthy generation in black-box LLM settings.

cs.CL↗