Searcharxiv⌕ Search

arXiv subjects

Jingyuan Wang

Publications and source records attributed to Jingyuan Wang.

At least 19 recordsLinked to original sources
arXiv2026

CityPlanner: A Sandbox Agent for Executable Urban Planning

Urban planning is a real-world spatial optimization problem that requires selecting feasible actions from large candidate spaces under practical objectives such as cost and service quality. Existing optimization and reinforcement learning methods are effective for fixed formulations, but often depend on task-specific representations and constraint handling. We propose \emph{CityPlanner}, a sandbox-agent framework for executable urban planning. CityPlanner introduces \emph{UrbanSandbox}, a unified file-based environment where agents inspect task files, generate plans, run evaluators, and revise decisions based on executable feedback. To make learning tractable, we further propose atomic-task reinforcement learning, which decomposes long sandbox trajectories into \emph{BuildPlan} for initial construction and \emph{ImprovePlan} for feedback-based refinement. Experiments on a real-world benchmark show that CityPlanner consistently outperforms heuristic, task-specific RL, and general LLM-agent baselines. Ablations verify the contributions of UrbanSandbox, atomic-task RL, and iterative deployment. We release the code and dataset at https://anonymous.4open.science/r/co-agent-C1C8

cs.AI↗
arXiv2026

Model Effect or Label Effect? Refined Annotations and a Human-Referenced Benchmark for Pulmonary Embolism Segmentation

Purpose: To quantify how evaluation annotations influence measured pulmonary embolism (PE) segmentation performance relative to model training changes, and to establish a human-referenced framework. Materials and Methods: This retrospective study screened 166 voxel-annotated CT pulmonary angiography cases from CADPE (n=91), FUMPE (n=35), and READ (n=40); 149 were included. A primary rater annotated PE by protocol, and a senior thoracic radiologist reviewed and revised all segmentations. Three additional raters at three centers annotated a 15-case subset. The label effect was measured by evaluating two pretrained nnU-Net models (nnU-Net-A, nnU-Net-B) against original and refined annotations. The model effect was measured by comparing the same architecture trained on different dataset combinations with annotations fixed. The benchmark model (nnPE) was trained with leave-one-dataset-out and pooled five-fold cross-validation. Four metric categories were analyzed with case-paired Wilcoxon signed-rank tests, Benjamini-Hochberg correction, and bootstrap 95% CIs. Results: Changing only the annotation increased mean DSC by 0.143 (0.122-0.166) for nnU-Net-A and 0.188 (0.163-0.213) for nnU-Net-B (both P < .001), whereas changing training-dataset composition changed DSC by 0.028. The label effect exceeded the model effect on CADPE and FUMPE and was 0.045 on READ. Within-mask attenuation SD fell in all three datasets after re-annotation (all P < .001). nnPE reached DSC 0.72 +/- 0.22 on pooled cross-validation but scored below all four annotators across 52 paired comparisons (all corrected P < .05). Conclusion: Evaluation annotations affected measured PE segmentation performance at least as much as model training choices. A human-referenced evaluation framework is publicly available for future study.

eess.IV↗
arXiv2026

A Modular Multitask Reasoning Framework Integrating Spatio-temporal Models and LLMs

Spatio-temporal data mining plays a pivotal role in informed decision making across diverse domains. However, existing models are often restricted to narrow tasks, lacking the capacity for multi-task inference and complex long-form reasoning that requires generation of in-depth, explanatory outputs. These limitations restrict their applicability to real-world, multi-faceted decision scenarios. In this work, we introduce STReason, a novel framework that integrates the reasoning strengths of large language models (LLMs) with the analytical capabilities of spatio-temporal models for multi-task inference and execution. Without task-specific fine-tuning, STReason leverages in-context learning to decompose complex natural language queries into modular, interpretable programs, which are then systematically executed to generate both numerical solutions and detailed reasoning rationales. By grounding all explanations in verified computational outputs, STReason inherently suppresses factual hallucinations common in direct-prompt LLM approaches. To enable rigorous evaluation, we construct a new benchmark dataset and propose a unified evaluation framework with metrics specifically designed for long-form spatio-temporal reasoning. Experimental results demonstrate that STReason significantly outperforms advanced LLM baselines across all reasoning metrics, particularly excelling in complex, reasoning-intensive spatio-temporal scenarios. Human evaluations further validate STReason's credibility and practical utility, demonstrating its potential to reduce expert workload and broaden the applicability to real-world spatio-temporal tasks.

cs.CL↗
arXiv2026

DEPT: Document Embedding Preservation Tuning for Unified Query Expansion and Retrieval

Large language models (LLMs) can both expand underspecified queries and encode text as dense representations, suggesting a unified model for query expansion and retrieval. Existing systems usually rely on prompted expansions, independently trained modules, or staged optimization, leaving generated expansions only indirectly aligned with the retrieval loss that judges them. We train a single decoder-only LLM end to end, where the same model generates the expansion and encodes both the expanded query and candidate documents. This unified setting creates a moving-target problem: retrieval supervision should improve query-side expansion, but the same update also shifts the document embeddings that serve as retrieval targets. We introduce Document Embedding Preservation Tuning (DEPT), which keeps tuned document embeddings close to cached initial embeddings while allowing retrieval gradients to pass through straight-through decoding into the generator. DEPT converts joint query--document movement into query-side adaptation against approximately stable, whitened document embeddings that support index reuse and online hard-negative mining. Experiments with Qwen3-4B-Instruct-2507 and LLaMA-3.2-3B-Instruct on five datasets in BEIR benchmark show that DEPT improves average retrieval quality over training-free, independently trained, and staged unified baselines, while ablations isolate the effects of preservation, whitening, end-to-end expansion training, and online negatives. Code is available at https://github.com/ILSparkle/DEPT.

cs.IR↗
arXiv2026

HieraMix: A Hierarchical MLP-Mixer for Large-Scale Traffic Forecasting

Traffic forecasting task is significant to modern urban management. Recently, there is growing attention on large-scale forecasting, as it better reflects the complexity of real-world traffic networks. However, existing models often exhibit quadratic computational complexity, making them impractical for large-scale real-world scenarios. In this paper, we propose a novel framework, Spatio-Temporal Hierarchical Mixer (HieraMix), which leverages an all-MLP architecture for efficient and effective large-scale traffic forecasting. HieraMix employs a hierarchical spatiotemporal mixing block to extract multi-resolution features through bottom-up aggregation and top-down propagation. Furthermore, an adaptive region mixer generates transformation matrices based on regional semantics, enabling our model to dynamically capture evolving spatiotemporal patterns for different regions. Extensive experiments conducted on four large-scale real-world datasets demonstrate that the proposed method not only achieves state-of-the-art performance but also exhibits competitive computational efficiency.

cs.LG↗
arXiv2026

Cosmological gravitational particle production in multifield inflation

We study cosmological gravitational particle production (CGPP) of dark matter in two-field inflationary backgrounds with both flat and curved field-space geometries. As a concrete realization of broader multifield mechanisms, we adopt a Starobinsky+quadratic potential and construct benchmark scenarios that interpolate between the flat field-space limit and the sidetracked attractor on a hyperbolic field space, and we compute the production spectrum of a gravitationally coupled spectator scalar for both minimal ($ξ= 0$) and conformal ($ξ= 1/6$) coupling to the Ricci scalar. We show that negative field-space curvature can strongly enhance the post-inflationary oscillations of the Ricci scalar, leading to an enhancement of up to an order of magnitude in the CGPP number density relative to the flat field-space limit, particularly for minimal coupling. For the sidetracked attractor, this enhanced production competes with a reduced inflationary energy scale, leading to a nontrivial dependence of the relic abundance on model parameters. We derive the relic abundance as a function of spectator mass and reheating temperature, and identify the viable parameter space for each benchmark. The conformal case $ξ= 1/6$, whose scalar mode equation is structurally analogous to that of a massive Dirac fermion, is much less constrained by isocurvature and provides a minimal scenario for purely gravitational dark matter production in multifield inflation.

hep-ph↗
arXiv2026

Automatic Attenuation Control for Mitigating Photon-Counting Saturation in SPAD-based Optical Wireless Communications

Single-photon avalanche diodes (SPADs) have emerged as a promising candidate for optical wireless communication (OWC) owing to their ultra-high sensitivity and singlephoton detection capability. However, under strong background radiation or high signal power, SPAD-based receivers suffer from photon-counting saturation, which severely degrades communication performance. To address this challenge, this paper introduces an automatic attenuation control (AAC) technique that dynamically optimizes the incident optical intensity to mitigate saturation effects. We develop a comprehensive analytical model for the SPAD-based OWC system, incorporating the influence of dead time and the lack of photon-number resolution. Based on this model, a convex optimization-based AAC algorithm is proposed to maximize the achievable rate in real time. Furthermore, a low-complexity AAC algorithm is devised using a closed-form trigger probability criterion, reducing computational complexity by two orders of magnitude. Numerical results demonstrate that the proposed AAC technique significantly improves both the achievable rate and symbol error rate across a wide range of background conditions, providing an efficient solution to enhance the dynamic range of photon-counting receivers.

eess.SP↗
arXiv2026

Unified Analytical Framework for SPAD Array Receivers with Dead-Time-Induced Blocking Loss and Inter-Symbol Interference in PAM-OWC Systems

Optical wireless communication (OWC) leveraging single-photon avalanche diode (SPAD) arrays offers exceptional sensitivity for photon-starving links. However, the inherent dead time of SPADs critically limits achievable data rates by introducing non-linear photon-counting distortions: blocking loss within a symbol duration and inter-symbol interference (ISI) across durations. This paper proposes a unified analytical framework capturing both distortions across all operational speed regimes for pulse-amplitude modulation (PAM), by establishing comprehensive statistical models for SPAD array receivers. For low and medium-speed systems (symbol duration longer than dead time), we derive exact closed-form expressions for the photon counts probability distribution using renewal theory, explicitly incorporating blocking loss and ISI. For high-speed systems (symbol duration shorter than dead time), we develop a Markov chain model characterizing the steady-state operational states and integrate it with trigger probability to obtain the exact binomial photon counts distribution. Furthermore, we propose low-complexity, near-optimal threshold detection schemes based on these models. This work provides essential theoretical tools for designing and optimizing high-performance SPAD-based OWC systems employing PAM.

eess.SP↗
arXiv2026

LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?

Large language models (LLMs) have demonstrated remarkable progress in reasoning, often through supervised fine-tuning (SFT). However, SFT is resource-intensive, relying on large curated datasets, rejection-sampled demonstrations, and uniform optimization across all tokens, even though only a fraction carry meaningful learning value. In this work, we explore a counterintuitive idea: can smaller language models (SLMs) teach larger language models (LLMs) by revealing high-value reasoning moments that reflect the latter's unique strength? We propose LightReasoner, a novel framework that leverages the behavioral divergence between a stronger expert model (LLM) and a weaker amateur model (SLM). LightReasoner operates in two stages: (1) a sampling stage that pinpoints critical reasoning moments and constructs supervision examples capturing the expert's advantage through expert-amateur contrast, and (2) a fine-tuning stage that aligns the expert model with these distilled examples, amplifying its reasoning strengths. Across seven mathematical benchmarks, LightReasoner improves accuracy by up to 28.1%, while reducing time consumption by 90%, sampled problems by 80%, and tuned token usage by 99%, all without relying on ground-truth labels. By turning weaker SLMs into effective teaching signals, LightReasoner offers a scalable and resource-efficient approach for advancing LLM reasoning. Code is available at: https://github.com/HKUDS/LightReasoner

cs.CL↗
arXiv2026

DLink: Distilling Layer-wise and Dominant Knowledge from EEG Foundation Models

EEG foundation models (EFMs) achieve strong cross-subject and cross-task generalization through large-scale pretraining and downstream fine-tuning. Through empirical analysis, we observe that (i) task-adapted EFMs provide strong decoding performance but incur substantial overhead when retained as inference backbones, making knowledge distillation a natural route for optimizing compact students; and (ii) direct distillation from a fixed teacher representation underutilizes EFM knowledge, as task-discriminative information is distributed across intermediate layers rather than concentrated in the final layer. These observations motivate DLink (Distilling Layer-wise and Dominant Knowledge), a spectrally guided distillation framework with input-conditioned layer routing for transferring EFM knowledge into compact students. DLink uses a lightweight router to aggregate teacher layers for each input, and aligns magnitude and phase spectra to mitigate compression-induced spectral distortion in learned representations. The routed teacher knowledge is internalized by a project-then-compress student; the teacher and router are used only during training. Experiments on four EEG benchmarks show that DLink improves matched compact students and remains competitive with lightweight baselines, narrowing the gap to fine-tuned EFMs while substantially reducing parameters, FLOPs, and CPU-only inference latency.

cs.LG↗
arXiv2026

SOAR: Real-Time Joint Optimization of Order Allocation and Robot Scheduling in Robotic Mobile Fulfillment Systems

Robotic Mobile Fulfillment Systems (RMFS) rely on mobile robots for automated inventory transportation, coordinating order allocation and robot scheduling to enhance warehousing efficiency. However, optimizing RMFS is challenging due to strict real-time constraints and the strong coupling of multi-phase decisions. Existing methods either decompose the problem into isolated sub-tasks to guarantee responsiveness at the cost of global optimality, or rely on computationally expensive global optimization models that are unsuitable for dynamic industrial environments. To bridge this gap, we propose SOAR, a unified Deep Reinforcement Learning framework for real-time joint optimization. SOAR transforms order allocation and robot scheduling into a unified process by utilizing soft order allocations as observations. We formulate this as an Event-Driven Markov Decision Process, enabling the agent to perform simultaneous scheduling in response to asynchronous system events. Technically, we employ a Heterogeneous Graph Transformer to encode the warehouse state and integrate phased domain knowledge. Additionally, we incorporate a reward shaping strategy to address sparse feedback in long-horizon tasks. Extensive experiments on synthetic and real-world industrial datasets, in collaboration with Geekplus, demonstrate that SOAR reduces global makespan by 7.5\% and average order completion time by 15.4\% with sub-100ms latency. Furthermore, sim-to-real deployment confirms its practical viability and significant performance gains in production environments. The code is available at https://github.com/200815147/SOAR.

cs.AI↗
arXiv2026

Creation of spin-3/2 dark matter via cosmological gravitational particle production

We study the cosmological gravitational particle production (CGPP) of spin-3/2 particles during and after cosmic inflation, and map the parameter space that can realize the observed dark matter density in stable spin-3/2 particles. Originally formulated by Rarita and Schwinger, the relativistic theory of a massive spin-3/2 field later found a home in supergravity as the superpartner of the graviton, and in nuclear physics as baryonic resonances and nuclear isotopes. We study a minimal model realization, namely a free massive spin-3/2 field minimally coupled to gravity, and adopt the name raritron for this field. We demonstrate that CGPP of raritrons crucially depends on the hierarchy between the raritron mass $m_{3/2}$ and the Hubble parameter at the end of inflation $H_e$, with high-mass and low-mass cases distinguished by the evolution of the sound speed $c_s$ of the longitudinal (helicity-1/2) mode, which is approximately unity at all times for heavy (relative to Hubble) raritrons and can become small or vanish for lighter raritrons, leading to a dramatic enhancement of production of high momentum particles in the latter case. Assuming the raritrons are stable, this leads to a wide parameter space to produce the observed dark matter density. Finally, we consider a time-dependent raritron mass, which can be chosen to remove the vanishing sound speed of the longitudinal mode, but which nonetheless enhances the production relative to the constant high-mass case, and in particular does not necessarily tame the high momentum tail of the spectrum. We perform our calculations using the Bogoliubov formalism and compare, when applicable, to the Boltzmann formalism.

hep-ph↗
arXiv2026

RoboClaw: An Agentic Framework for Scalable Long-Horizon Robotic Tasks

Vision-Language-Action (VLA) systems have shown strong potential for language-driven robotic manipulation. However, scaling them to long-horizon tasks remains challenging. Existing pipelines typically separate data collection, policy learning, and deployment, resulting in heavy reliance on manual environment resets and brittle multi-policy execution. We present RoboClaw, an agentic robotics framework that unifies data collection, policy learning, and task execution under a single VLM-driven controller. At the policy level, RoboClaw introduces Entangled Action Pairs (EAP), which couple forward manipulation behaviors with inverse recovery actions to form self-resetting loops for autonomous data collection. This mechanism enables continuous on-policy data acquisition and iterative policy refinement with minimal human intervention. During deployment, the same agent performs high-level reasoning and dynamically orchestrates learned policy primitives to accomplish long-horizon tasks. By maintaining consistent contextual semantics across collection and execution, RoboClaw reduces mismatch between the two phases and improves multi-policy robustness. Experiments in real-world manipulation tasks demonstrate improved stability and scalability compared to conventional open-loop pipelines, while significantly reducing human effort throughout the robot lifecycle, achieving a 25% improvement in success rate over baseline methods on long-horizon tasks and reducing human time investment by 53.7%.

cs.RO↗
arXiv2026

Reconciling strange metal transport in CeCoIn$_5$ through the difference of optical and cyclotron effective masses

The strange metal behavior in cuprate superconductors - characterized by linear in temperature resistivity and anomalous Hall transport - stands in stark contrast to the expectation of conventional Fermi liquid (FL) theory. Remarkably, the similar transport behavior has also been observed in the heavy fermion metal CeCoIn$_5$, whose d-wave superconducting ground state and strong antiferromagnetic fluctuations draw parallels to the cuprates. Here we have investigated the optical conductivity of the strange metal state of CeCoIn$_5$ over a wide magnetic field range using time-domain THz spectroscopy (TDTS). Using unique high-field THz spectroscopy we have shown that the current relaxation rate scales approximately as T$^2$, giving evidence for a hidden Fermi liquid state over a large field range. This result can be reconciled with linear in T resistivity with the realization that heavy quasiparticles have an optical mass that scales roughly like 1/T. This optical mass contrasts with the mass that characterizes cyclotron motion, which does not suffer the same large temperature dependent renormalization. Although by itself anomalous, this allows one to understand a number of other phenomena in CeCoIn$_5$ that have been taken to be signatures of strange metals, including the coexistence of a conventional T$^2$ dependence of the cotangent of the Hall angle with the linear in T resistivity, which with our observation also reflects FL-like physics.

cond-mat.str-el↗
arXiv2026

OSInsert: Towards High-authenticity and High-fidelity Image Composition

Generative image composition aims to regenerate the given foreground object in the background image to produce a realistic composite image. Some high-authenticity methods can adjust foreground pose/view to be compatible with background, while some high-fidelity methods can preserve the foreground details accurately. However, existing methods can hardly achieve both goals at the same time. In this work, we propose a two-stage strategy to achieve both goals. In the first stage, we use high-authenticity method to generate reasonable foreground shape, serving as the condition of high-fidelity method in the second stage. The experiments on MureCOM dataset verify the effectiveness of our two-stage strategy. The code and model have been released at https://github.com/bcmi/OSInsert-Image-Composition.

cs.CV↗
arXiv2026

UniPhy: Unifying Riemannian-Clifford Geometry and Biorthogonal Dynamics for Planetary-Scale Continuous Weather Modeling

While data-driven weather models have achieved remarkable deterministic accuracy, they fundamentally rely on discrete-time mappings and closed-system assumptions, failing to capture the multi-scale continuous dynamics and thermodynamic openness of the atmosphere. To address these limitations, we propose UniPhy, a continuous-time non-Hermitian neural stochastic partial differential equation (SPDE) solver. Geometrically, we employ Riemannian-Clifford gauge transformations to flatten planetary heterogeneity, enabling globally consistent operations. Dynamically, we construct non-Hermitian biorthogonal spectral operators integrated with a global flux tracker to capture transient energy growth and open-system exchange. Computationally, by identifying the algebraic associativity of the analytic solution, we reformulate adaptive physical integration as a parallel prefix-sum problem, achieving log-linear sequence parallelism. UniPhy establishes a physically complete foundation model architecture that unifies geometric adaptivity, thermodynamic consistency, and computational efficiency. Our code is available at .

physics.comp-ph↗
arXiv2026

Genie Centurion: Accelerating Scalable Real-World Robot Training with Human Rewind-and-Refine Guidance

While Vision-Language-Action (VLA) models show strong generalizability in various tasks, real-world deployment of robotic policy still requires large-scale, high-quality human expert demonstrations. However, data collection via human teleoperation requires continuous operator attention, which is costly, hard to scale. To address this, we propose Genie Centurion (GCENT), a scalable and general data collection paradigm based on human rewind-and-refine guidance, enabling robots' interactive learning in deployment. GCENT starts at an imperfect policy and improves over time. When the robot execution failures occur, GCENT allows robots to revert to a previous state with a rewind mechanism, after which a teleoperator provides corrective demonstrations to refine the policy. This framework supports a one-human-to-many-robots supervision scheme with a Task Sentinel module, which autonomously predicts task success and solicits human intervention when necessary. Empirical results show that GCENT achieves up to 40% higher task success rates than state-of-the-art data collection methods, and reaches comparable performance using less than half the data in long-horizon and precise tasks. We also quantify the data yield-to-effort ratio under multi-robot scenarios, demonstrating GCENT's potential for scalable and cost-efficient robot policy training in real-world environments.

cs.RO↗
arXiv2025

Multi-Agent Pointer Transformer: Seq-to-Seq Reinforcement Learning for Multi-Vehicle Dynamic Pickup-Delivery Problems

This paper addresses the cooperative Multi-Vehicle Dynamic Pickup and Delivery Problem with Stochastic Requests (MVDPDPSR) and proposes an end-to-end centralized decision-making framework based on sequence-to-sequence, named Multi-Agent Pointer Transformer (MAPT). MVDPDPSR is an extension of the vehicle routing problem and a spatio-temporal system optimization problem, widely applied in scenarios such as on-demand delivery. Classical operations research methods face bottlenecks in computational complexity and time efficiency when handling large-scale dynamic problems. Although existing reinforcement learning methods have achieved some progress, they still encounter several challenges: 1) Independent decoding across multiple vehicles fails to model joint action distributions; 2) The feature extraction network struggles to capture inter-entity relationships; 3) The joint action space is exponentially large. To address these issues, we designed the MAPT framework, which employs a Transformer Encoder to extract entity representations, combines a Transformer Decoder with a Pointer Network to generate joint action sequences in an AutoRegressive manner, and introduces a Relation-Aware Attention module to capture inter-entity relationships. Additionally, we guide the model's decision-making using informative priors to facilitate effective exploration. Experiments on 8 datasets demonstrate that MAPT significantly outperforms existing baseline methods in terms of performance and exhibits substantial computational time advantages compared to classical operations research methods.

cs.LG↗