arXiv subjects

Pei Li

Publications and source records attributed to Pei Li.

At least 19 recordsLinked to original sources

arXiv2026

Why We Created Yet Another Memory Framework: Understanding MGA's Role in Next-Gen Database Systems

Despite the presence of multiple memory regions in modern database systems, supporting an efficient form of memory remains a challenge under production constraints. In enterprise-grade data systems, existing abstractions impose a trade-off between coarse-grained global sharing and strict process isolation, resulting in data copying, memory fragmentation, and limited support for controlled sharing. These challenges become more pronounced as workloads grow more diverse, and systems must tolerate process failures while maintaining predictable performance. This paper introduces the Managed Global Area (MGA), a scoped shared-memory abstraction in Oracle AI Database that addresses these limitations. MGA allows components to explicitly define allocation source, membership, and coordination semantics across selected processes while integrating with a production database engine. Unlike fully shared memory regions in Oracle, such as the System Global Area (SGA), MGA supports dynamic process membership and modular memory usage without imposing system-wide visibility. We evaluate MGA on analytical and AI workloads that stress shared-memory execution, including TPC-H hash joins and ONNX Runtime inference. Under concurrent execution, MGA reduces latency for join-intensive TPC-H queries by up to 35%. For ONNX- based inference, MGA-enabled model sharing reduces memory footprint by up to 90% and lowers large-model inference latency by up to 37%. These results demonstrate that dynamically scoped shared memory can improve both efficiency and predictability in production database systems.

cs.DB↗

arXiv2026

FakeI2V-Bench: Benchmarking the Applicability of Image-level Deepfake Detectors for Deepfake Video Detection

Recent advances in video generation models have significantly intensified the deepfake threat, yet the current deepfake video detection benchmarks remain underdeveloped. In particular, the effectiveness of image-level detectors in the video domain has not been systematically assessed. To fill this gap, we present FakeI2V-Bench, a benchmark for evaluating state-of-the-art video-level deepfake detectors in challenging scenarios, with a particular focus on systematically assessing the performance of image-level deepfake detectors in the video domain. FakeI2V-Bench comprises 97,548 videos, containing content generated by the latest powerful generation models and covering a broader range of categories. Using this dataset, we conduct a systematic evaluation of eight video-level detectors and twelve representative image-level detectors. Experimental results show that the best-performing image-level detector achieves an 80.16% AUC, slightly outperforming the strongest video-level detector (i.e., 79.99% AUC). Going beyond benchmarking, we present IV-Bridge, a general framework that enhances the applicability of image-level deepfake detectors to videos. IV-Bridge employs a random forest model with statistical features to aggregate frame-level predictions, allowing eleven image-level detectors to surpass state-of-the-art video-level approaches, with the best-performing variant achieving a 93.80% AUC. Overall, FakeI2V-Bench establishes a rigorous benchmark for deepfake video detection and introduces a novel pathway for extending image-level detectors to the video domain, offering new insights and directions for future research. Code and data are available at https://github.com/CryptoAILab/FakeI2V-Bench.

cs.CR↗

arXiv2026

Proton-proton Femtoscopy as a Probe of Short-range Structure in High-Energy O+O Collisions

Short-range nucleon-nucleon correlations are a defining feature of the nuclear many-body wave function, yet they are invisible in the one-body density and therefore inaccessible to observables that measure a nuclear size. We show that proton-proton femtoscopy supplies the missing sub-femtometer sensitivity. In $^{16}$O+$^{16}$O collisions at $\rm \sqrt{s_{NN}}=$ 200 GeV, we compare three nuclear-structure inputs spanning mean-field, low-resolution cluster, and short-range-correlated descriptions. The $p$-$p$ correlation function separates all three, most sharply in peripheral collisions, where the \textit{ab initio} input suppresses the extracted source radius by $\sim5\%$ relative to the mean-field baseline. Under identical conditions $\pi^{+}$-$\pi^{+}$ correlations respond an order of magnitude more weakly, and the $C_{pp}/C_{\pi^{+}\pi^{+}}$ double ratio retains the full effect, pointing to the short-distance weighting of the $^{1}S_{0}$ pair rather than to an overall rescaling of the source. The signal survives the leading theoretical systematic, the choice of strong-interaction potential, which we quantify explicitly. These results identify $p$-$p$ femtoscopy as a short-distance-resolved probe of light-nucleus structure, complementary to flow observables that constrain only the low-order moments of the initial geometry.

nucl-th↗

arXiv2026

Schr\"{o}dinger Generator for High-Dimensional Integration and Sampling on Quantum Many-Body States

Integration and sampling in high dimensions are among central challenges in modern science and technology, underlying applications ranging from quantum many-body physics to Bayesian inference and artificial intelligence. Although conventional Monte Carlo methods are formally scalable, their efficiency deteriorates rapidly in the presence of strong correlations or sharp features in high-dimensional configuration space. Here we introduce a new framework, termed the Schr"odinger Generator, for integration and sampling based on the explicit optimization of coordinate transformations. The method decomposes the total Jacobian into two complementary components, including an adaptive map that minimizes estimator variance by learning the marginal structure in each dimension, and a normalizing-flow-based transformation that captures non-factorizable correlations in the target distribution. A final resampling step guarantees unbiased sampling even when the learned transformation is imperfect. We demonstrate stable and scalable performance for nuclear quantum many-body states in dimensions exceeding 600. Short-range correlations among nucleons in finite nucleus are faithfully reproduced. The framework offers a physically transparent approach to high-dimensional stochastic integration and sampling, opening new possibilities for simulations of complex quantum systems.

nucl-th↗

arXiv2026

Robust Watermarks Meet Backdoored Models: Evading Diffusion Semantic Watermarks via Stealthy Backdoor

Although semantic watermarking is considered a promising safeguard for images generated by Latent Diffusion Models (LDMs), the reliance of the watermark detection pipeline on neural networks introduces a critical yet underexplored backdoor attack surface. To systematically study this vulnerability, we propose GhostVAE to plant a stealthy backdoor into the encoder of Variational Autoencoder (VAE), enabling reliable evasion of watermark detection. GhostVAE operates in two stages: it first constructs a universal trigger via power spectrum regularization to improve the trigger robustness, and then trains a backdoored VAE encoder with a parameter-aligned objective. Through extensive evaluations across three state-of-the-art semantic watermarking schemes and three widely adopted LDMs, we show that GhostVAE preserves watermark detection performance on benign images (achieving an average true positive rate of 94.4%), while simultaneously enabling highly effective evasion under trigger activation (achieving an average attack success rate of 94.6%). Moreover, we comprehensively analyze seventeen representative defenses and demonstrate that GhostVAE remains stealthy across the input space, parameter space, and latent space. Our work fundamentally undermines the trustworthiness of semantic watermarking systems and highlights that secure deployment of semantic watermarks requires end-to-end security considerations, particularly for neural network components.

cs.CR↗

arXiv2026

Forecasting the Emergence and Evolution of Crash Hotspots: A Unified Deep Learning Framework for Proactive Traffic Safety

Road crashes remain among the gravest threats to public safety, and preventing them is a defining task of transportation systems worldwide. Much of that harm concentrates at hotspots, yet a hotspot is less a place than an episode; it emerges quietly at an intersection or along an arterial, intensifies for weeks, then subsides, only to reappear elsewhere. Enforcement guided by maps of past crashes inevitably trails this cycle, patrolling yesterday's hotspots while tomorrow's form unwatched. Breaking that lag requires three capabilities at once: detecting hotspots as they are born, forecasting where they will sit next week, and following each one through its life. We introduce HERALD (Hotspot Emergence, Risk Anticipation, and Life-cycle Dynamics), a unified deep learning framework that provides all three from a single statewide model. HERALD distills each county's recent crash history into weekly risk maps and forecasts the next with a CNN--Transformer, whose mixture-of-experts lets one model serve dense urban cores and sparse rural corridors alike. Each forecast is anchored in the county's long-run crash geography, sharpened by the self-exciting effect of recent crashes, and paired with explicit warnings of where new hotspots are about to appear. Followed over time, every hotspot acquires a legible life story, from birth through growth and stability to decline and death. Across six heterogeneous Wisconsin counties, HERALD forecasts more accurately than five identically trained baselines, locates hotspots most precisely, and flags emerging risks before they take hold. A single adjustable setting trades accuracy for extra sensitivity where deployment demands it. The result shifts hotspot management from mapping the past to anticipating the future.

cs.LG↗

arXiv2026

Geometry-Aware Infrastructure-Anchored Denoiser for UWB Sensing and Work-Zone Reconstruction

Accurate work-zone geometry perception is critical for intelligent transportation systems, and ultra-wideband sensing offers a low-cost approach for infrastructure-aided reconstruction. However, outdoor UWB ranging is often degraded by non-line-of-sight propagation, burst noise, and long-tail errors, which can distort downstream spatial reconstruction. We present GAIA, a geometry-aware, infrastructure-anchored learning framework that couples temporal range modeling with latent anchor-layout estimation and deterministic distance projection. GAIA preserves range denoising as the supervised task while orienting the learned distances toward boundary-consistent reconstruction. We evaluate GAIA on a real-world outdoor UWB dataset with synchronized UWB, GNSS, and IMU measurements, and further test robustness using a real-data-calibrated stress-test simulator. GAIA achieves the lowest overall range MSE and highest polygon IoU among evaluated filtering-based and learning-based baselines, reducing MSE by 18.4% and improving polygon IoU by 15.5% over PoseMLP. These results show that geometry-aware range denoising provides an effective path toward spatially coherent work-zone reconstruction.

cs.LG↗

arXiv2026

YouZhi: Towards High-Concurrency Financial LLMs via Adaptive GQA-to-MLA Transition

Large language models (LLMs) drive significant financial innovations, yet their high-concurrency deployment is severely bottlenecked by KV cache memory overhead, which inflates infrastructure costs and throttles scalability. To address this, we propose YouZhi-LLM, a highly efficient financial LLM empowered by a comprehensive structural transition and training pipeline natively built on the Huawei Ascend ecosystem. At its algorithmic core, YouZhi-LLM features a layer-adaptive GQA-to-MLA transition framework that dynamically assigns per-layer FreqFold sizes, maximizing KV-cache compression while minimizing perplexity degradation. To recover representation capacity and inject domain expertise, the Ascend-based training pipeline seamlessly integrates generalized knowledge distillation with financial-specific supervised fine-tuning. Evaluations demonstrate the superiority of this systematic approach, with the adaptive transition reducing perplexity degradation by up to 35% over uniform baselines. Crucially, when evaluated on Ascend NPUs via vLLM-Ascend, the massive KV-cache reduction translates directly into deployment efficiency. Compared to their respective base models, YouZhi-7B yields a 12.3% improvement in average financial benchmark score alongside a 2.69$\times$ increase in maximum concurrency; similarly, YouZhi-14B achieves a 7.0% accuracy gain and a 2.43$\times$ concurrency boost, establishing a new paradigm for cost-effective, high-throughput financial inference.

cs.CL↗

arXiv2026

Scalable All-Optical Fibre-Mode Data Transmission with Profiles-Preserved Decoding

Optical fibres are the primary medium for optical signal transmission, and their guided modes provide a high-dimensional basis for modal-domain information encoding. However, conventional demultiplexing approaches typically convert fibre modes into fundamental Gaussian modes and require repeated mode conversions, while existing profiles-preserved methods are generally restricted to fewer than three modes. High-quality fibre-mode data transmission therefore requires a scalable all-optical decoder capable of separating strongly overlapping modal channels while preserving their intrinsic spatial profiles. Here, we establish a scalable profiles-preserved all-optical decoding method for high-dimensional fibre-mode data transmission. By introducing a microlens-array-assisted decoding architecture with channel-dependent spherical phase compensation, the proposed method accommodates mode-dependent effective focal-length variations, enabling scalable modal-channel separation while preserving high-quality modal profiles at the output plane. Experimentally, the optical decoder resolved fields containing eight fibre modes, achieving a mode fidelity exceeding 0.72, a worst-channel crosstalk of $-5.57~\mathrm{dB}$ and a mean non-target crosstalk of $-21.34~\mathrm{dB}$, while reconstructing the relative modal weights with an error below 0.1. Semantic transmission experiments using digits and Chinese characters further demonstrated effective recovery of the encoded information from the decoded modal signals. We expect this work to provide a scalable route towards high-dimensional all-optical fibre-mode data transmission.

physics.optics↗

arXiv2026

Multiple fidelities and joint numerical range

We investigate the effectiveness of entanglement detection based on multiple fidelities via the geometry of the joint separable numerical range. When all reference states are product states, we derive a necessary and sufficient criterion for such detection: either some pair of reference states has nontrivial moduli of the local inner products on both subsystems, or the orthogonal complement of the span of the reference states is completely entangled. We further show that there exist sets of reference product states for which no proper subset is effective for entanglement detection, whereas the full set is. A typical example of this phenomenon is provided by unextendible product bases. Moreover, for a pair of reference product states on a bipartite system with arbitrary local dimensions, we characterize both the joint numerical range and the joint separable numerical range, showing that the joint separable numerical range is determined solely by their local fidelities, as illustrated by a representative two-qubit example. Our results offer a systematic approach to designing effective entanglement witnesses and lay the groundwork for extensions to higher-dimensional and multipartite scenarios.

quant-ph↗

arXiv2026

Behavior-Grounded Lane Representation Learning for Multi-Task Traffic Digital Twins

Traffic digital twins are powerful tools for advanced traffic management, and most systems are built on static geometric representations. However, these representations fail to capture the dynamic functional semantics required for behavior-aware reasoning, such as how a lane operates under complex traffic conditions. To address this gap, we introduce GeoLaneRep, a behavior-grounded lane representation learning framework for traffic digital twins. GeoLaneRep jointly encodes static lane geometry, observed vehicle trajectories, and operational descriptors into a shared, cross-camera semantic embedding. The encoder is trained with a joint objective combining contrastive cross-camera alignment, auxiliary role supervision, and temporal anomaly detection. Across 16 roadside cameras and 132 lanes, the learned embeddings achieve a $0.004$ lateral-rank error and an edge-role F1 of $1.000$ in zero-shot cross-camera matching, and an AUROC of $0.991$ for window-level anomaly detection. We further show that the same behavioral embeddings can condition a diffusion-based generator to synthesize lane geometries that satisfy targeted operational specifications, with $87.9\%$ overall specification accuracy across 38 lane groups. GeoLaneRep thus provides a semantic interface between roadside observations and downstream digital twin tasks, supporting cross-camera transfer, behavior-aware monitoring, and goal-directed lane synthesis. The framework is openly available at https://github.com/raynbowy23/GeoLaneRep.

cs.CV↗

arXiv2026

An Agentic Workflow for Detecting Personally Identifiable Information in Crash Narratives

Crash narratives in crash reports provide crucial contextual information for traffic safety analysis. Yet, their broader use is hindered by the presence of personally identifiable information (PII), including names, home addresses, and license plate numbers. Because PII appears sparsely and inconsistently in crash narratives, manual detection is not scalable, and existing rule-based approaches often fail to capture context-dependent PII. This study develops and evaluates a locally deployable, agentic workflow for PII detection in crash narratives by leveraging large language models (LLMs). The workflow contains a Hybrid Extractor and a Verifier. The Hybrid Extractor routes structured PII (e.g., phone numbers and email addresses) to a rule-based model (i.e., Presidio) and context-dependent PII (e.g., names, home addresses, and alphanumeric identifiers) to a domain-adapted, fine-tuned LLM. To address ambiguity in challenging categories, the workflow incorporates ensemble LLM extraction and an agentic verification step that filters false detections through evidence-based reasoning. Evaluated on a real-world crash dataset, the agentic workflow achieves strong performance with a precision of 0.82, a recall of 0.94, an F1 of 0.87, and an accuracy of 0.96, outperforming multiple baseline methods. Moreover, the ablation results suggest that ensemble LLM extraction and Verifier offer improved detection for home addresses and alphanumeric identifiers. The workflow runs locally, supporting privacy-sensitive operational settings where external APIs are restricted. This work offers a practical and robust path for scalable, privacy-preserving crash data processing, enabling broader research and safety interventions while safeguarding individual privacy.

cs.CR↗

arXiv2026

CrashSight: A Phase-Aware, Infrastructure-Centric Video Benchmark for Traffic Crash Scene Understanding and Reasoning

Cooperative autonomous driving requires traffic scene understanding from both vehicle and infrastructure perspectives. While vision-language models (VLMs) show strong general reasoning capabilities, their performance in safety-critical traffic scenarios remains insufficiently evaluated due to the ego-vehicle focus of existing benchmarks. To bridge this gap, we present \textbf{CrashSight}, a large-scale vision-language benchmark for roadway crash understanding using real-world roadside camera data. The dataset comprises 250 crash videos, annotated with 13K multiple-choice question-answer pairs organized under a two-tier taxonomy. Tier 1 evaluates the visual grounding of scene context and involved parties, while Tier 2 probes higher-level reasoning, including crash mechanics, causal attribution, temporal progression, and post-crash outcomes. We benchmark 8 state-of-the-art VLMs and show that, despite strong scene description capabilities, current models struggle with temporal and causal reasoning in safety-critical scenarios. We provide a detailed analysis of failure scenarios and discuss directions for improving VLM crash understanding. The benchmark provides a standardized evaluation framework for infrastructure-assisted perception in cooperative autonomous driving. The CrashSight benchmark, including the full dataset and code, is accessible at https://mcgrche.github.io/crashsight.

cs.CV↗

arXiv2026

V2X-QA: A Comprehensive Reasoning Dataset and Benchmark for Multimodal Large Language Models in Autonomous Driving Across Ego, Infrastructure, and Cooperative Views

Multimodal large language models (MLLMs) have shown strong potential for autonomous driving, yet existing benchmarks remain largely ego-centric and therefore cannot systematically assess model performance in infrastructure-centric and cooperative driving conditions. In this work, we introduce V2X-QA, a real-world dataset and benchmark for evaluating MLLMs across vehicle-side, infrastructure-side, and cooperative viewpoints. V2X-QA is built around a view-decoupled evaluation protocol that enables controlled comparison under vehicle-only, infrastructure-only, and cooperative driving conditions within a unified multiple-choice question answering (MCQA) framework. The benchmark is organized into a twelve-task taxonomy spanning perception, prediction, and reasoning and planning, and is constructed through expert-verified MCQA annotation to enable fine-grained diagnosis of viewpoint-dependent capabilities. Benchmark results across ten representative state-of-the-art proprietary and open-source models show that viewpoint accessibility substantially affects performance, and infrastructure-side reasoning supports meaningful macroscopic traffic understanding. Results also indicate that cooperative reasoning remains challenging since it requires cross-view alignment and evidence integration rather than simply additional visual input. To address these challenges, we introduce V2X-MoE, a benchmark-aligned baseline with explicit view routing and viewpoint-specific LoRA experts. The strong performance of V2X-MoE further suggests that explicit viewpoint specialization is a promising direction for multi-view reasoning in autonomous driving. Overall, V2X-QA provides a foundation for studying multi-perspective reasoning, reliability, and cooperative physical intelligence in connected autonomous driving. The dataset and V2X-MoE resources are publicly available at: https://github.com/junwei0001/V2X-QA.

cs.RO↗

arXiv2026

Order Is Not Layout: Order-to-Space Bias in Image Generation

We study a systematic bias in modern image generation models: the mention order of entities in text spuriously determines spatial layout and entity--role binding. We term this phenomenon Order-to-Space Bias (OTS) and show that it arises in both text-to-image and image-to-image generation, often overriding grounded cues and causing incorrect layouts or swapped assignments. To quantify OTS, we introduce OTS-Bench, which isolates order effects with paired prompts differing only in entity order and evaluates models along two dimensions: homogenization and correctness. Experiments show that Order-to-Space Bias (OTS) is widespread in modern image generation models, and provide evidence that it is primarily data-driven and manifests during the early stages of layout formation. Motivated by this insight, we show that both targeted fine-tuning and early-stage intervention strategies can substantially reduce OTS, while preserving generation quality.

cs.CL↗

arXiv2026

WS-IMUBench: Can Weakly Supervised Methods from Audio, Image, and Video Be Adapted for IMU-based Temporal Action Localization?

IMU-based Human Activity Recognition (HAR) has enabled a wide range of ubiquitous computing applications, yet its dominant clip classification paradigm cannot capture the rich temporal structure of real-world behaviors. This motivates a shift toward IMU Temporal Action Localization (IMU-TAL), which predicts both action categories and their start/end times in continuous streams. However, current progress is strongly bottlenecked by the need for dense, frame-level boundary annotations, which are costly and difficult to scale. To address this bottleneck, we introduce WS-IMUBench, a systematic benchmark study of weakly supervised IMU-TAL (WS-IMU-TAL) under only sequence-level labels. Rather than proposing a new localization algorithm, we evaluate how well established weakly supervised localization paradigms from audio, image, and video transfer to IMU-TAL under only sequence-level labels. We benchmark seven representative weakly supervised methods on seven public IMU datasets, resulting in over 3,540 model training runs and 7,080 inference evaluations. Guided by three research questions on transferability, effectiveness, and insights, our findings show that (i) transfer is modality-dependent, with temporal-domain methods generally more stable than image-derived proposal-based approaches; (ii) weak supervision can be competitive on favorable datasets (e.g., with longer actions and higher-dimensional sensing); and (iii) dominant failure modes arise from short actions, temporal ambiguity, and proposal quality. Finally, we outline concrete directions for advancing WS-IMU-TAL (e.g., IMU-specific proposal generation, boundary-aware objectives, and stronger temporal reasoning). Beyond individual results, WS-IMUBench establishes a reproducible benchmarking template, datasets, protocols, and analyses, to accelerate community-wide progress toward scalable WS-IMU-TAL.

cs.CV↗

arXiv2025

Universal imprinting of short-range correlations in relativistic heavy-ion collisions

Protons and neutrons within atomic nuclei undergo intense and fleeting encounters driven by the strong force at short distances. These interactions generate close-proximity pairs, a phenomenon known as short-range correlations (SRCs). While the properties of SRCs have been extensively studied in cold nuclear matter, their behavior under extremely hot and dense conditions remains largely unexplored. Here, we incorporate correlated nucleon configurations into relativistic heavy-ion collisions, using the quark-gluon plasma (QGP), a state of matter present microseconds after the Big Bang, as a sensitive diagnostic tool. We find that these correlations induce substantial modifications to event-by-event geometry, which are quantitatively identified through higher-order moments of the transverse profile. Most importantly, we find a surprising linear relation between QGP geometry fluctuations and the SRC scale factor spanning systems from deuteron to lead, which reflects the universal imprinting of SRCs in relativistic heavy-ion collisions. Our findings reveal the emergence of short-range structural effects across vastly different energy scales from low-energy electron scattering to high-energy nuclear collisions.

nucl-th↗

arXiv2025

Real-Time Lane-Level Crash Detection on Freeways Using Sparse Telematics Data

Real-time traffic crash detection is critical in intelligent transportation systems because traditional crash notifications often suffer delays and lack specific, lane-level location information, which can lead to safety risks and economic losses. This paper proposes a real-time, lane-level crash detection approach for freeways that only leverages sparse telematics trajectory data. In the offline stage, the historical trajectories are discretized into spatial cells using vector cross-product techniques, and then used to estimate a vehicle intention distribution and select an alert threshold by maximizing the F1-score based on official crash reports. In the online stage, incoming telematics records are mapped to these cells and scored for three modules: transition anomalies, speed deviations, and lateral maneuver risks, with scores accumulated into a cell-specific risk map. When any cell's risk exceeds the alert threshold, the system issues a prompt warning. Relying solely on telematics data, this real-time and low-cost solution is evaluated on a Wisconsin dataset and validated against official crash reports, achieving a 75% crash identification rate with accurate lane-level localization, an overall accuracy of 96%, an F1-score of 0.84, and a non-crash-to-crash misclassification rate of only 0.6%, while also detecting 13% of crashes more than 3 minutes before the recorded crash time.

eess.SY↗