Searcharxiv⌕ Search

arXiv subjects

Lei Wei

Publications and source records attributed to Lei Wei.

At least 19 recordsLinked to original sources
arXiv2026

JPO: Juris Policy Optimization for Structured Legal Reasoning in Criminal Judgment Prediction

Criminal judgment prediction requires models to infer statutory articles, charges, and sentencing outcomes from case facts. Unlike standard classification tasks, it involves a structured reasoning process in which statutes should be matched with facts, charges should be justified by statutes, and sentencing outcomes should remain consistent with charges. Existing approaches optimize final labels, and while some have attempted to evaluate reasoning quality, their evaluations are indirect, often relying on LLM-generated rubrics that reflect model-internal preferences rather than the inherent logical structure of legal adjudication. We propose Juris Policy Optimization (JPO), a post-training framework for structured legal reasoning in Chinese criminal judgment prediction. JPO first uses teacher-generated rationales to supervise a standardized four-step reasoning process, and then applies reinforcement learning with a composite reward over legal prediction quality, reasoning structure completeness, and cross-step consistency. JPO further introduces token-level advantage reweighting and adaptive clipping for legally salient reasoning segments. Experiments on multiple open-source language models and three Chinese legal benchmarks show that JPO consistently improves both judgment prediction and reasoning quality over supervised fine-tuning and reinforcement learning baselines.

cs.CL↗
arXiv2026

Modality Fault Lines: Structural Corruptions Reveal Fragile Omni-Modal Reasoning

Omni-modal large language models are increasingly evaluated on clean text--vision--audio inputs, where every channel is present, synchronized, and readily interpretable. Such scores are often taken as evidence of robust cross-modal fusion, but clean evaluation cannot tell whether success depends on stable cross-modal structure or on cues sufficient only in intact inputs. To address this gap, we define a modality fault line: a boundary at which model behavior becomes unstable when a modality remains present and human-interpretable, but its internal evidence structure is perturbed. We introduce SCEval (Structure-Corruption Evaluation) a diagnostic evaluation protocol that keeps the question, answer space, and modality channels fixed while applying controlled structural corruptions to text, vision, and audio individually and jointly. Built from $273$ human-verified tri-modal examples from Social-IQ, OmniBench, and VALOR, SCEval evaluates $15$ proprietary and open-source omni-modal systems. The results show that structural corruption lowers clean accuracy, text--vision damage forms the most stable shared fault line, and multi-modal degradation is non-additive rather than a simple function of the number of corrupted modalities. Clean omni-modal accuracy therefore does not establish that a model will remain reliable when cross-modal evidence becomes structurally unreliable.

cs.CL↗
arXiv2026

Dispersive decay for the Nonlinear magnetic Schr\"{o}dinger equation

In this paper, we obtain the dispersive estimates and global well-posedness of the nonlinear magnetic Schr\"odinger equation in $\mathbb{R}^{3}$ with nonlinearity $|u|^{p-1}u$ with exponent $\frac{5}{3}< p<5$ when the initial value stays in a suitable space $\Sigma_{s}$. By proving the resolvent estimates with weight functions and the \textquotedblleft almost equivalence\textquotedblright between $(-\Delta_{A})^{\frac{s}{2}}$ and $(-\Delta)^{\frac{s}{2}}$, we obtain the Strichartz estimates of $|J_{A}(t)|^{s}u:=e^{\frac{i|x|^{2}}{4t}}(-t^{2}\Delta_{A})^{\frac{s}{2}}e^{\frac{-i|x|^{2}}{4t}}u$, therefore the dispersive decay estimates are obtained.

math.SP↗
arXiv2026

Plans You Can Check: Verifier-Grounded Learning of an Open-Weight Planner for Executable Video-Editing

Practical video editing is not only pixel generation: an editor must turn a brief, a clip pool, music metadata, and hard constraints into an executable timeline. We study this decision layer as \emph{executable video-editing planning} and introduce RefineCut, which, unlike workflow systems that wrap a prompted frontier model, trains a compact open-weight planner for it. The planner edits a typed timeline through structured patches covering clip selection, trimming, ordering, transitions, and duration and music alignment; a deterministic verifier applies each patch and checks it against an explicit constraint ledger. Because editing has no single ground-truth repair, we do not imitate teachers directly: RefineCut replays every multi-teacher branch through the verifier and keeps verifier-best repairs as supervision. A second stage, RefineCut-Evo, lets the student score its own repairs with the verifier and a task rubric and trains on high-margin preference pairs, so the final $8$B planner runs in a closed verifier loop with no teacher calls at inference. On RefineCut-Bench ($3{,}578$ tasks, $7{,}971$ captioned clips, $499$ music tracks, explicit ledgers), verifier-replayed distillation lifts the planner from $0.620$ to $0.858$ on the protocol-specific Video-Editing Score and RefineCut-Evo reaches $0.924$; the gain transfers to Llama-3.1-8B and GLM-4-9B, and in the same closed loop the $8$B planner matches or exceeds its frontier teachers. Code and RefineCut-Bench are publicly released; see the Data Availability statement.

cs.CV↗
arXiv2026

COMET: Contrastive Motion-Enhanced Temporal Reasoning for Video Multimodal Large Language Models

Video multimodal large language models have advanced significantly, yet fine-grained motion-temporal understanding remains fragile. The core bottleneck is not only sparse frame sampling, but also the lack of a complete temporal modeling pipeline for explicitly representing frame-to-frame change, enabling appearance-motion interaction, and optimizing temporal direction sensitivity. We propose COMET, a temporally grounded framework that systematically strengthens video MLLMs through explicit temporal representation, appearance-motion fusion, and direction-aware optimization. Architecturally, COMET introduces a temporal motion branch built on Taylor frame differences and injects its motion evidence into the appearance stream via temporal attention bias-enhanced cross-attention. For optimization, COMET combines temporal prior distillation with a forward-reverse TC-GRPO stage that turns temporal order into a direct learning signal and strengthens the model's use of directional motion patterns encoded by the temporal motion branch. The method achieves consistent overall improvements with a pronounced motion-temporal bias: on Qwen3-VL-8B, action-centric tasks (STAR, SSv2) improve by 4.9% on average, temporal reasoning tasks (NExT-QA, CLEVRER, LLaVA-178K) by 2.1% over BL-GRPO, while static perception tasks (PerceptionTest) remain on par. The same gain pattern also transfers to InternVL2.5-8B, indicating that COMET generalizes across model families.

cs.CV↗
arXiv2026

SafeFlow: Semantic Information-Flow Control for Blocking Malicious Propagation in Multi-Agent Systems

Multi-agent systems improve capability through task decomposition and role specialization, but these same mechanisms introduce an important safety blind spot: a harmful objective can be fragmented into locally plausible subtasks, allowing malicious intent to evade detection by any single agent. This is a growing social-impact challenge: systems handling sensitive information or consequential tools can turn routine delegation into unauthorized disclosure or unsafe action. We argue that this failure mode is better understood as a semantic information-flow problem than as a single-turn prompt classification task. To address this, we propose SafeFlow, a defense framework for multi-agent systems that formalizes malicious cross-agent propagation as a semantic information-flow problem. SafeFlow attaches structured semantic taints to root requests, propagates them through a dynamic collaboration graph, and performs workflow-level validation to reconstruct the global risk context before irreversible actions are committed. Evaluated on four benchmarks spanning prompt injection, jailbreak-based unsafe tool use, risky code execution, and harmful web-agent behavior, SafeFlow reduces attack success rates compared to undefended baselines and external defenses while retaining high benign task completion and a high paired safe--harm success rate. Our findings show that multi-agent systems still lack mechanisms for preserving risk semantics across delegation boundaries. This gap can turn routine delegation into privacy harms or unsafe actions that affect people and organizations. SafeFlow keeps this risk visible throughout the workflow, before it results in harm.

cs.MA↗
arXiv2026

Dynamic Speed Limit Control of Connected Automated Vehicles in Freeway Networks Considering Traffic Composition Uncertainty

Dynamic speed limit control has emerged as a promising strategy to improve freeway sustainability in mixed traffic environments with connected automated vehicles (CAVs). However, most existing approaches assume that the CAV penetration rate is deterministic and can be accurately known throughout the control horizon. In reality, the penetration rate has inherent observation errors, leading to uncertainty in mixed traffic composition, which in turn degrades control performance. To overcome this limitation, this study proposes a novel model predictive control (MPC) framework for dynamic CAV speed limit control in freeway networks that explicitly incorporates traffic composition uncertainty into both flow prediction and control optimization. An uncertainty-aware macroscopic mixed traffic model is first developed, where the uncertain penetration rate propagates through the mixed fundamental diagram to the flow dynamics by affecting the mixed free-flow speed, capacity, and capacity drop condition. Then, a traffic composition-aware MPC is formulated to optimize CAV speed limits against multiple admissible penetration rate realizations, thereby improving control robustness under heterogeneous traffic conditions. Simulation experiments are conducted on both a single-bottleneck freeway corridor and a multi-bottleneck freeway network with merge-diverge interactions. The results demonstrate that the proposed controller generates more spatially coordinated speed limits, which effectively reduce travel time spent and provide environmental benefits.

eess.SY↗
arXiv2026

Link-Based Multimodal Traffic Dynamics Model in Continuous-Time Framework

Computationally efficient models for multimodal traffic flows with inter-modal interactions are foundational for coordinated traffic management. However, such models are lacking in the current literature. This study introduces a multimodal link transmission model (M-LTM) accommodating both continuous road traffic and discrete tramway traffic at the network level. M-LTM builds on a link model that captures the inter-modal interactions through the moving bottleneck theory. We further develop node models describing the flow transfer, tram operations, and mode interactions at tram stops and intersections. Specifically, the tram dwell process and dwell-induced congestion of road traffic are simulated at the stop node, and the intersection node model can reproduce queue spillback and tram priority. Case studies are conducted on two synthetic one-node networks, a synthetic arterial network, and the real network of Dresden to demonstrate the predictive power and the applicability of the proposed model in network traffic analysis and management. The GEH statistics of road traffic are below 4, and the average tram arrival time absolute errors are less than 10 seconds.

math.OC↗
arXiv2026

A comparative and critical study of EEGNet for fNIRS-driven cognitive load classification

Accurately classifying cognitive load from functional near-infrared spectroscopy (fNIRS) signals remains a significant challenge due to temporal variability, inter-subject differences, and sensitivity to preprocessing choices. This study provides a comprehensive evaluation of EEGNet for fNIRS-based cognitive load classification by systematically examining the effects of temporal segmentation strategies (overlapping vs. non-overlapping), window lengths (10s, 20s, 30s), feature extraction methods (Analysis of Variance (ANOVA), Principal Component Analysis (PCA), Fast Independent Component Analysis (FastICA)), learning rate configurations (fixed and adaptive), and evaluation protocols (random split vs. subject-independent (SI)). Results from random-split experiments show that overlapping segmentation, combined with smaller fixed learning rates (0.01-0.001), yields the highest accuracies, due to temporal redundancy and dense sampling of hemodynamic transitions. However, SI evaluation reveals a substantial drop in accuracy, demonstrating limited generalization to unseen participants. Under SI evaluation, non-overlapping segmentation outperformed overlapping windows, with the best accuracy of 56.11% achieved using PCA features with a 20-second window and a 0.1 learning rate. These findings indicate that eliminating temporal redundancy helps the model learn more robust and generalizable representations of cognitive load across individuals. Although adaptive learning rate strategy improved training stability, it did not surpass the performance of optimally selected fixed learning rates. The study highlights the critical role of segmentation strategy and learning rate selection in improving model generalization and identifies methodological considerations essential for developing reliable, real-time, and SI cognitive load classification systems using fNIRS.

cs.LG↗
arXiv2026

Room-temperature tuning and probing of Fermi polarons in atomically thin semiconductors on a plasmonic metasurface

The Fermi polaron, arising from interactions between a mobile impurity and a degenerate Fermi sea, is a many-body quasiparticle that provides a sensitive probe of strongly correlated electronic phases in atomically thin semiconductors. In doped transition-metal dichalcogenides, the attractive and repulsive polaron branches are well established in monolayers. However, extending active control and quantitative, branch-resolved probing to stacked geometries has remained elusive because spectral quenching and weak optical contrast restrict access to Fermi polaron signatures. Here, we integrate electron-doped WS$_2$ flakes from monolayer to quadrilayer with a strain-tunable plasmonic metasurface, enabling high-contrast scattering readout at room temperature through coupling between Fermi polaron resonances and surface plasmons. This platform enables quantitative extraction of polaron branch spectral weights and coupling strengths across different layer numbers. We uncover a systematic thickness dependence of the spectral-weight distribution and demonstrate continuous and fully reversible spectral-weight transfer between attractive and repulsive branches in bilayers and quadrilayers, with near-complete transfer achieved in bilayers. By identifying layer number and strain as complementary control parameters for Fermi polarons, our results establish metasurface-enabled scattering spectroscopy as a practical route to resolve and manipulate many-body resonances in stacked van der Waals semiconductors, bridging idealized monolayer polaron physics and device-relevant architectures.

cond-mat.mes-hall↗
arXiv2026

Out of Sight, Not Out of Mind: Unveiling Latent Attack in Latent-based Multi-Agent Systems

Latent-based multi-agent systems replace parts of explicit inter-agent communication with hidden representations, offering a new direction for efficient and flexible agent collaboration. However, moving coordination into latent space may also move attacks beyond the reach of visible-text inspection. In this paper, we study whether latent states can carry attack-associated information that remains effective during clean executions. To examine this question, we introduce a latent attack framework that reactivates attack-induced effects through latent interventions without reusing adversarial text. Extensive experiments show that the resulting latent attacks can substantially degrade task performance in clean executions, especially when applied to inter-agent KV-cache handoffs rather than local hidden states. Further control analyses indicate that this degradation cannot be reduced to arbitrary perturbations or invalid generation. Overall, our findings suggest that latent-based collaboration does not remove attack risk. It shifts part of the risk into less observable execution states, calling for safeguards beyond visible-text inspection.

cs.CR↗
arXiv2026

General Hazard Detection

Hazard, as an abstract concept, is typically defined through cognitive-level logical reasoning rather than concrete examples. In contrast, existing hazard detection systems rely on predefined hazard categories and require intensive collection of labelled examples within detection or classification architectures. This approach faces three fundamental challenges when addressing abstract safety concepts: (1) noisy and sparse training data, (2) dynamically evolving definitions that change across contexts and time, and (3) limited generalisation to unseen or novel scenarios. To address these limitations, we present the CompliVision dataset, the first general-purpose hazard dataset designed for rule-based compliance assessment, along with a baseline framework for hazard evaluation. Our key innovation is decoupling the hazard concept from image-based examples by expressing safety requirements through language-based rules. We ground our approach in authoritative domain regulations and ISO standards to define diverse hazard concepts across multiple domains. The CompliVision dataset comprises 3,006 images spanning traffic, construction, and warehouse environments, with each image annotated for compliance against specific safety rules, accompanied by natural language explanations highlighting the supporting visual evidence. To achieve robust generalisation, we develop an active learning framework to more effectively guide and refine vision-language models in assessing hazard compliance. While state-of-the-art VLMs demonstrate strong capabilities, they struggle with the fine-grained, context-dependent interpretation required for accurate safety assessment. We proposed a general hazard detection framework to address this limitation which combines LLaVA-based visual reasoning with with human-in-the-loop feedback.

cs.CV↗
arXiv2026

An Object-Centered Data Acquisition Method for 3D Gaussian Splatting using Mobile Phones

Data acquisition through mobile phones remains a challenge for 3D Gaussian Splatting (3DGS). In this work we target the object-centered scenario and enable reliable mobile acquisition by providing on-device capture guidance and recording onboard sensor signals for offline reconstruction. After the calibration step, the device orientations are aligned to a baseline frame to obtain relative poses, and the optical axis of the camera is mapped to an object-centered spherical grid for uniform viewpoint indexing. To curb polar sampling bias, we compute area-weighted spherical coverage in real-time and guide the user's motion accordingly. We compare the proposed method with RealityScan and the free-capture strategy. Our method achieves superior reconstruction quality using fewer input images compared to free capture and RealityScan. Further analysis shows that the proposed method is able to obtain more comprehensive and uniform viewpoint coverage during object-centered acquisition.

cs.CV↗
arXiv2026

From Explanations to Architecture: Explainability-Driven CNN Refinement for Brain Tumor Classification in MRI

Recent brain tumor classification methods often report high accuracy but rely on deep, over-parameterized architectures with limited interpretability, making it difficult to determine whether predictions are driven by tumor-relevant evidence or by spurious cues such as background artifacts or normal tissue. We propose an explainable convolutional neural network (CNN) framework that enhances model transparency without sacrificing classification accuracy. This approach supports more trustworthy AI in healthcare and contributes to SDG 3: Good Health and Well-being by enabling more dependable MRI-based brain tumor diagnosis and earlier detection. Rather than using explainable AI solely for post hoc visualization, we employ Grad-CAM to quantify layer-wise relevance and guide the removal of low-contribution layers, reducing unnecessary depth and parameters while encouraging attention to discriminative tumor regions. We further validate the model's decision rationale using complementary explainability methods, combining Grad-CAM for spatial localization with SHAP and LIME for attribution-based verification. Experiments on multi-class brain MRI datasets show that the proposed model achieves 98.21% accuracy on the primary dataset and 95.74% accuracy on an unseen dataset, indicating strong cross-dataset generalization. Overall, the proposed approach balances simplicity, transparency, and accuracy, supporting more trustworthy and clinically applicable brain tumor classification for improved health outcomes and non-invasive disease detection.

eess.IV↗
arXiv2026

How did the Urban Network Flow Adapt to the Collapse of the Carola Bridge?

The unexpected collapse of the Carola Bridge in Dresden, Germany, provides a rare opportunity to characterise how urban network traffic adapts to an unexpected infrastructure disruption. This study develops a data-driven analytical framework using traffic data from the Dresden traffic management system to assess the short-term impacts of the disruption. By combining statistical comparisons of pre- and post-collapse motorised traffic distributions, peak-hour shifts, and Park-and-Ride data analyses, the framework reveals how traffic dynamics and traveller choices adjust under infrastructure disruption. Results reveal that the two closest bridges, the Albert and Marien Bridges, absorb the majority of the diverted motorised traffic. In particular, the daily traffic volume on the Albert bridge increases by up to 81%, which is equivalent to 3.5 hours of traffic operating with maximum flow. Peak hours on critical links are significantly prolonged, reaching up to 250 minutes. Besides redistribution, the overall daily motorised traffic crossing the Elbe river declines by approximately 8,000 vehicles, while Park-and-Ride usage increases by up to 188%, suggesting a potential travel mode shift after the disruption. The study reveals the patterns of traffic redistribution following an unexpected disruption and provides insights for resilience planning and emergency traffic management.

nlin.AO↗
arXiv2026

Think-Augmented Function Calling: Improving LLM Parameter Accuracy Through Embedded Reasoning

Large language models (LLMs) have demonstrated remarkable capabilities in function calling for autonomous agents, yet current mechanisms lack explicit reasoning transparency during parameter generation, particularly for complex functions with interdependent parameters. While existing approaches like chain-of-thought prompting operate at the agent level, they fail to provide fine-grained reasoning guidance for individual function parameters. To address these limitations, we propose Think-Augmented Function Calling (TAFC), a novel framework that enhances function calling accuracy through explicit reasoning at both function and parameter levels. Our method introduces a universal "think" parameter augmentation that enables models to articulate their decision-making process, with dynamic optimization for parameter descriptions to improve reasoning quality. For complex parameters, TAFC automatically triggers granular reasoning based on complexity scoring, ensuring appropriate justification for critical decisions. Additionally, we propose reasoning-guided optimization to align generated reasoning with human expectations. TAFC requires no architectural modifications to existing LLMs while maintaining full API compatibility. Evaluation on ToolBench across proprietary and open-source models demonstrates significant improvements in parameter generation accuracy and reasoning coherence for multi-parameter functions, while providing enhanced interpretability for debugging AI agent behaviors.

cs.AI↗
arXiv2026

FadeMem: Biologically-Inspired Forgetting for Efficient Agent Memory

Large language models deployed as autonomous agents face critical memory limitations, lacking selective forgetting mechanisms that lead to either catastrophic forgetting at context boundaries or information overload within them. While human memory naturally balances retention and forgetting through adaptive decay processes, current AI systems employ binary retention strategies that preserve everything or lose it entirely. We propose FadeMem, a biologically-inspired agent memory architecture that incorporates active forgetting mechanisms mirroring human cognitive efficiency. FadeMem implements differential decay rates across a dual-layer memory hierarchy, where retention is governed by adaptive exponential decay functions modulated by semantic relevance, access frequency, and temporal patterns. Through LLM-guided conflict resolution and intelligent memory fusion, our system consolidates related information while allowing irrelevant details to fade. Experiments on Multi-Session Chat, LoCoMo, and LTI-Bench demonstrate superior multi-hop reasoning and retrieval with 45\% storage reduction, validating the effectiveness of biologically-inspired forgetting in agent memory systems.

cs.AI↗
arXiv2026

JADE: Expert-Grounded Dynamic Evaluation for Open-Ended Professional Tasks

Evaluating agentic AI on open-ended professional tasks faces a fundamental dilemma between rigor and flexibility. Static rubrics provide rigorous, reproducible assessment but fail to accommodate diverse valid response strategies, while LLM-as-a-judge approaches adapt to individual responses yet suffer from instability and bias. Human experts address this dilemma by combining domain-grounded principles with dynamic, claim-level assessment. Inspired by this process, we propose JADE, a two-layer evaluation framework. Layer 1 encodes expert knowledge as a predefined set of evaluation skills, providing stable evaluation criteria. Layer 2 performs report-specific, claim-level evaluation to flexibly assess diverse reasoning strategies, with evidence-dependency gating to invalidate conclusions built on refuted claims. Experiments on BizBench show that JADE improves evaluation stability and reveals critical agent failure modes missed by holistic LLM-based evaluators. We further demonstrate strong alignment with expert-authored rubrics and effective transfer to HealthBench and DR.BENCH, covering medical and 10-domain professional evaluation settings. Code and data are available at https://github.com/smiling-world/JADE.

cs.AI↗