arXiv subjects

Lu Zhang

Publications and source records attributed to Lu Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2026

Toward Functional and Non-Functional Evaluation of Application-Level Code Generation

Large language models (LLMs) have achieved strong performance on code generation. However, most prior evaluations focus on snippet-level outputs, such as function generation or repository completion. These settings do not fully evaluate application-level code generation, where the goal is to produce a runnable repository with coherent multi-file structure, dependency support, and end-to-end executability. In addition, real-world software quality depends not only on functional correctness but also on non-functional quality attributes, such as maintainability and security. In this paper, we present RAL-Bench, a benchmark and evaluation framework for application-level code generation. For each task, RAL-Bench derives a concise natural-language requirement from a high-quality reference project, constructs black-box system tests for both functional correctness and non-functional quality attributes. It also retains only the candidate tests that pass on the reference repository. Under this unified evaluation protocol, functional correctness is measured by the system test pass rate, while non-functional quality is evaluated along five ISO/IEC 25010-inspired dimensions, with per-dimension diagnostics and reference-normalized scoring.We evaluate 16 frontier LLMs under a controlled zero-shot setting with greedy decoding. The results show that functional correctness remains the primary bottleneck in application-level code generation, while non-functional quality also remains challenging. Under our evaluation protocol, no model exceeds a 45\% functional score. These findings suggest that strong performance on existing code generation benchmarks does not yet translate to strong performance on application-level repository generation. This result highlights the need for evaluation settings that directly assess end-to-end repository generation rather than relying only on snippet-level success.

cs.SE↗

arXiv2026

A scalable and programmable optical neural network in a time-synthetic dimension

Programmable optical neural networks (ONNs) can offer high-throughput and energy-efficient solutions for accelerating artificial intelligence (AI) computing. However, existing ONN architectures, typically based on cascaded unitary transformations such as Mach-Zehnder interferometer meshes, face inherent scalability limitations due to spatial encoding, which causes optical components and system complexity to scale quadratically with network size. A promising solution to this challenge is the use of synthetic dimensions to enhance scalability, though experimental demonstration has remained scarce. Here, we present the first experimental demonstration of an all-optical, highly scalable, programmable ONN operating in a time-synthetic dimension. By implementing a time-cycle computation paradigm analogous to gate cycling in conventional spatial photonic circuits, our approach achieves a gate count surpassing that of state-of-the-art programmable photonic processors. Unlike conventional ONN architectures that rely on real-space wave interferences, our framework exploits time-reflection and time-refraction to perform computations, fundamentally eliminating backscattering errors through causality constraints. To bridge the gap between simulation and reality, we introduce an in-situ training framework that dynamically adapts to experimental errors, achieving performance exceeding traditional in silico learning paradigms. Our synthetic-dimension-based approach provides a compact, scalable, backscattering-free, and programmable neuromorphic computing architecture, advancing the potential for next-generation photonic AI systems.

physics.optics↗

arXiv2026

Reference-guided Policy Optimization for Molecular Optimization via LLM Reasoning

Large language models (LLMs) benefit substantially from supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR) in reasoning tasks. However, these recipes perform poorly in instruction-based molecular optimization, where each data point typically provides only a single optimized reference molecule and no step-by-step optimization trajectory. We reveal that answer-only SFT on the reference molecules collapses reasoning, and RLVR provides sparse feedback under similarity constraints due to the model's lack of effective exploration, which slows learning and limits optimization. To encourage the exploration of new molecules while balancing the exploitation of the reference molecules, we introduce Reference-guided Policy Optimization (RePO), an optimization approach that learns from reference molecules without requiring trajectory data. At each update, RePO samples candidate molecules with their intermediate reasoning trajectories from the model and trains the model using verifiable rewards that measure property satisfaction under similarity constraints in an RL manner. Meanwhile, it applies reference guidance by keeping the policy's intermediate reasoning trajectory as context and training only the answer in a supervised manner. Together, the RL term promotes exploration, while the guidance term mitigates reward sparsity and stabilizes training by grounding outputs to references when many valid molecular edits exist. Across molecular optimization benchmarks, RePO consistently outperforms SFT and RLVR baselines (e.g., GRPO), achieving improvements on the optimization metric (Success Rate $\times$ Similarity), improving balance across competing objectives, and generalizing better to unseen instruction styles. Our code is publicly available at https://github.com/tmlr-group/RePO.

cs.LG↗

arXiv2026

UETrack: A Unified and Efficient Framework for Single Object Tracking

With growing real-world demands, efficient tracking has received increasing attention. However, most existing methods are limited to RGB inputs and struggle in multi-modal scenarios. Moreover, current multi-modal tracking approaches typically use complex designs, making them too heavy and slow for resource-constrained deployment. To tackle these limitations, we propose UETrack, an efficient framework for single object tracking. UETrack demonstrates high practicality and versatility, efficiently handling multiple modalities including RGB, Depth, Thermal, Event, and Language, and addresses the gap in efficient multi-modal tracking. It introduces two key components: a Token-Pooling-based Mixture-of-Experts mechanism that enhances modeling capacity through feature aggregation and expert specialization, and a Target-aware Adaptive Distillation strategy that selectively performs distillation based on sample characteristics, reducing redundant supervision and improving performance. Extensive experiments on 12 benchmarks across 3 hardware platforms show that UETrack achieves a superior speed-accuracy trade-off compared to previous methods. For instance, UETrack-B achieves 69.2% AUC on LaSOT and runs at 163/56/60 FPS on GPU/CPU/AGX, demonstrating strong practicality and versatility. Code is available at https://github.com/kangben258/UETrack.

cs.CV↗

arXiv2026

Phases and criticality of the triangular lattice SU(N) Hofstadter-Hubbard model

We report the study of phases and transitions of SU(N) Hofstadter-Hubbard model subject to commensurate magnetic field on the triangular lattice. At filling one fermion per site, for the number of fermion flavors 2 <= N <= 8, we identify three distinct phases and calculate critical interaction strength from parton large-N mean-field approximation. Integer quantum Hall, chiral spin liquid, and valence bond solid states could be realized upon varying the Hubbard interaction U and the number of flavor N . We construct the critical theory for the putative continuous transition from quantum Hall states to chiral spin liquid and calculate the critical transport behavior using quantum Boltzmann equations for general N . These results could be validated in synthetic systems such as moire superlattices and cold atom platforms.

cond-mat.str-el↗

arXiv2026

GA-Field: Geometry-Aware Vehicle Aerodynamic Field Prediction

Accurate aerodynamic field prediction is crucial for vehicle drag evaluation, but the computational cost of high-fidelity CFD hinders its use in iterative design workflows. While learning-based methods enable fast and scalable inference, accurately aerodynamic fields modeling remains challenging, as it demands capturing both long-range geometric effects and fine-scale flow structures. Existing approaches typically encode geometry only once at the input and formulate prediction as a one-shot mapping, which often leads to diluted global shape awareness and insufficient resolution of sharp local flow variations. To address these issues, we propose GA-Field, a Geometry-Aware Field prediction network that introduces two complementary design components: (i) a global geometry injection mechanism that repeatedly conditions the network on a compact 3D geometry embedding at multiple stages to preserve long-range geometric consistency, and (ii) a coarse-to-fine field refinement strategy to recover sharp local aerodynamic details. GA-Field achieves new state-of-the-art performance on ShapeNet-Car and the large-scale DrivAerNet++ benchmark for surface pressure, wall shear stress, and 3D velocity prediction tasks, while exhibiting strong out-of-distribution generalization across different vehicle categories.

cs.CE↗

arXiv2026

Disentangled Representation Learning for Parametric Partial Differential Equations

Neural operators (NOs) excel at learning mappings between function spaces, serving as efficient forward solution approximators for PDE-governed systems. However, as black-box solvers, they offer limited insight into the underlying physical mechanism, due to the lack of interpretable representations of the physical parameters that drive the system. To tackle this challenge, we propose a new paradigm for learning disentangled representations from NO parameters, thereby effectively solving an inverse problem. Specifically, we introduce DisentangO, a novel hyper-neural operator architecture designed to unveil and disentangle latent physical factors of variation embedded within the black-box neural operator parameters. At the core of DisentangO is a multi-task NO architecture that distills the varying parameters of the governing PDE through a task-wise adaptive layer, alongside a variational autoencoder that disentangles these variations into identifiable latent factors. By learning these disentangled representations, DisentangO not only enhances physical interpretability but also enables more robust generalization across diverse systems. Empirical evaluations across supervised, semi-supervised, and unsupervised learning contexts show that DisentangO effectively extracts meaningful and interpretable latent features, bridging the gap between predictive performance and physical understanding in neural operator frameworks.

cs.LG↗

arXiv2026

Quasi-multimodal-based pathophysiological feature learning for retinal disease diagnosis

Retinal diseases spanning a broad spectrum can be effectively identified and diagnosed using complementary signals from multimodal data. However, multimodal diagnosis in ophthalmic practice is typically challenged in terms of data heterogeneity, potential invasiveness, registration complexity, and so on. As such, a unified framework that integrates multimodal data synthesis and fusion is proposed for retinal disease classification and grading. Specifically, the synthesized multimodal data incorporates fundus fluorescein angiography (FFA), multispectral imaging (MSI), and saliency maps that emphasize latent lesions as well as optic disc/cup regions. Parallel models are independently trained to learn modality-specific representations that capture cross-pathophysiological signatures. These features are then adaptively calibrated within and across modalities to perform information pruning and flexible integration according to downstream tasks. The proposed learning system is thoroughly interpreted through visualizations in both image and feature spaces. Extensive experiments on two public datasets demonstrated the superiority of our approach over state-of-the-art ones in the tasks of multi-label classification (F1-score: 0.683, AUC: 0.953) and diabetic retinopathy grading (Accuracy:0.842, Kappa: 0.861). This work not only enhances the accuracy and efficiency of retinal disease screening but also offers a scalable framework for data augmentation across various medical imaging modalities.

cs.CV↗

arXiv2026

Community-Level Modeling of Gyral Folding Patterns for Robust and Anatomically Informed Individualized Brain Mapping

Cortical folding exhibits substantial inter-individual variability while preserving stable anatomical landmarks that enable fine-scale characterization of cortical organization. Among these, the three-hinge gyrus (3HG) serves as a key folding primitive, showing consistent topology yet meaningful variations in morphology, connectivity, and function. Existing landmark-based methods typically model each 3HG independently, ignoring that 3HGs form higher-order folding communities that capture mesoscale structure. This simplification weakens anatomical representation and makes one-to-one matching sensitive to positional variability and noise. We propose a spectral graph representation learning framework that models community-level folding units rather than isolated landmarks. Each 3HG is encoded using a dual-profile representation combining surface topology and structural connectivity. Subject-specific spectral clustering identifies coherent folding communities, followed by topological refinement to preserve anatomical continuity. For cross-subject correspondence, we introduce Joint Morphological-Geometric Matching, jointly optimizing geometric and morphometric similarity. Across over 1000 Human Connectome Project subjects, the resulting communities show reduced morphometric variance, stronger modular organization, improved hemispheric consistency, and superior alignment compared with atlas-based and landmark-based or embedding-based baselines. These findings demonstrate that community-level modeling provides a robust and anatomically grounded framework for individualized cortical characterization and reliable cross-subject correspondence.

q-bio.NC↗

arXiv2026

Parametric Hyperbolic Conservation Laws: A Unified Framework for Conservation, Entropy Stability, and Hyperbolicity

We propose a parametric hyperbolic conservation law (SymCLaw) for learning hyperbolic systems directly from data while ensuring conservation, entropy stability, and hyperbolicity by design. Unlike existing approaches that typically enforce only conservation or rely on prior knowledge of the governing equations, our method parameterizes the flux functions in a form that guarantees real eigenvalues and complete eigenvectors of the flux Jacobian, thereby preserving hyperbolicity. At the same time, we embed entropy-stable design principles by jointly learning a convex entropy function and its associated flux potential, ensuring entropy dissipation and the selection of physically admissible weak solutions. A corresponding entropy-stable numerical flux scheme provides compatibility with standard discretizations, allowing seamless integration into classical solvers. Numerical experiments on benchmark problems, including Burgers, shallow water, Euler, and KPP equations, demonstrate that SymCLaw generalizes to unseen initial conditions, maintains stability under noisy training data, and achieves accurate long-time predictions, highlighting its potential as a principled foundation for data-driven modeling of hyperbolic conservation laws.

math.NA↗

arXiv2026

Domain-Expert-Guided Hybrid Mixture-of-Experts for Medical AI: Integrating Data-Driven Learning with Clinical Priors

Mixture-of-Experts (MoE) models increase representational capacity with modest computational cost, but their effectiveness in specialized domains such as medicine is limited by small datasets. In contrast, clinical practice offers rich expert knowledge, such as physician gaze patterns and diagnostic heuristics, that models cannot reliably learn from limited data. Combining data-driven experts, which capture novel patterns, with domain-expert-guided experts, which encode accumulated clinical insights, provides complementary strengths for robust and clinically meaningful learning. To this end, we propose Domain-Knowledge-Guided Hybrid MoE (DKGH-MoE), a plug-and-play and interpretable module that unifies data-driven learning with domain expertise. DKGH-MoE integrates a data-driven MoE to extract novel features from raw imaging data, and a domain-expert-guided MoE incorporates clinical priors, specifically clinician eye-gaze cues, to emphasize regions of high diagnostic relevance. By integrating domain expert insights with data-driven features, DKGH-MoE improves both performance and interpretability.

cs.CV↗

arXiv2026

Do We Really Need SFT? Prompt-as-Policy over Knowledge Graphs for Cold-start Next POI Recommendation

Next point-of-interest (POI) recommendation is a key component of smart urban services, yet it remains challenging under cold-start conditions with sparse user-POI interactions. Recent LLM-based methods address this issue through either supervised fine-tuning (SFT) or in-context learning (ICL), but SFT is costly and prone to overfitting active users, while static prompts in ICL lack adaptability to diverse user contexts. We argue that the main limitation lies not in LLM reasoning ability, but in how contextual evidence is constructed and presented. Accordingly, we propose Prompt-as-Policy over knowledge graphs (KG), a reinforcement-guided prompting framework that formulates prompt construction as a learnable decision process, while keeping the LLM frozen as a reasoning engine. To enable structured prompt optimization, we organize heterogeneous user-POI signals into a KG and transform mined relational paths into evidence cards, which serve as atomic semantic units for prompt composition. A contextual bandit learner then optimizes a prompt policy that adaptively determines (i) which relational evidences to include, (ii) how many evidences to retain per candidate POI, and (iii) how to organize and order them within the prompt. Experiments on three real-world datasets show that Prompt-as-Policy consistently outperforms state-of-the-art baselines, achieving an average 11.87% relative improvement in Acc@1 for inactive users, while maintaining competitive performance for active users, without any model fine-tuning.

cs.SI↗

arXiv2026

Gyral-Sulcal-Net: An Integrated Network Representation of Brain Folding Patterns

Our brain functions as a complex communication network, and studying it from a network perspective offers valuable insights into its organizational principles and links to cognitive functions and brain disorders. However, most current network studies typically use brain regions as nodes, often overlooking the intricate folding patterns of finer-scale anatomical landmarks within these regions. In this study, we introduce a novel approach to integrate the brain's two primary folding patterns - gyri and sulci - into a unified network termed the Gyral-Sulcal-Net (GS-Net), in which three different types of finer-scale landmarks have been successfully identified. We evaluated the proposed GS-Net across multiple datasets, comprising over 1,600 brains, spanning different age groups (from 34 gestational weeks to elderly adults) and cohorts (healthy brains and those with pathological conditions). The experimental results demonstrate that the GS-Net can effectively integrate and represent diverse cortical folding patterns from a network perspective. More importantly, this approach offers a promising way for integrating different folding patterns into a unified anatomical brain network, alongside structural and functional networks, providing a comprehensive framework for studying brain networks.

q-bio.NC↗

arXiv2026

Machine learning aided parameter analysis in Perovskite X-ray Detector

Many factors in perovskite X-ray detectors, such as crystal lattice and carrier dynamics, determine the final device performance (e.g., sensitivity and detection limit). However, the relationship between these factors remains unknown due to the complexity of the material. In this study, we employ machine learning to reveal the relationship between 15 intrinsic properties of halide perovskite materials and their device performance. We construct a database of X-ray detectors for the training of machine learning. The results show that the band gap is mainly influenced by the atomic number of the B-site metal, and the lattice length parameter b has the greatest impact on the carrier mobility-lifetime product (μτ). An X-ray detector (m-F-PEA)2PbI4 were generated in the experiment and it further verified the accuracy of our ML models. We suggest further study on random forest regression for X-ray detector applications.

cond-mat.mtrl-sci↗

arXiv2026

Coupling Deep Learning with Full Waveform Inversion

Full waveform inversion (FWI) aims to reconstruct unknown physical coefficients in wave equations using the wavefield data generated from multiple incoming sources. In this work, we propose an offline-online computational strategy for coupling classical least-squares-based computational inversion with modern learning-based approaches for FWI to achieve advantages that can not be achieved with only one of the components. \RED{In brief, we develop an offline learning strategy to construct a robust approximation of the inverse operator through weighted optimization and utilize it to design a new objective function for approximate online inversion with new datasets. The approximate online inversion then serves as a warm start for the true online inversion.} We demonstrate through numerical simulations that our coupling strategy improves the computational efficiency of FWI with reliable offline training on moderate computational resources (in terms of both the size of the training dataset and the computational cost needed).

math.NA↗

arXiv2026

Towards Cross-Platform Generalization: Domain Adaptive 3D Detection with Augmentation and Pseudo-Labeling

This technical report represents the award-winning solution to the Cross-platform 3D Object Detection task in the RoboSense2025 Challenge. Our approach is built upon PVRCNN++, an efficient 3D object detection framework that effectively integrates point-based and voxel-based features. On top of this foundation, we improve cross-platform generalization by narrowing domain gaps through tailored data augmentation and a self-training strategy with pseudo-labels. These enhancements enabled our approach to secure the 3rd place in the challenge, achieving a 3D AP of 62.67% for the Car category on the phase-1 target domain, and 58.76% and 49.81% for Car and Pedestrian categories respectively on the phase-2 target domain.

cs.CV↗

arXiv2026

Breaking Model Lock-in: Cost-Efficient Zero-Shot LLM Routing via a Universal Latent Space

The rapid proliferation of Large Language Models (LLMs) has led to a fragmented and inefficient ecosystem, a state of ``model lock-in'' where seamlessly integrating novel models remains a significant bottleneck. Current routing frameworks require exhaustive, costly retraining, hindering scalability and adaptability. We introduce ZeroRouter, a new paradigm for LLM routing that breaks this lock-in. Our approach is founded on a universal latent space, a model-agnostic representation of query difficulty that fundamentally decouples the characterization of a query from the profiling of a model. This allows for zero-shot onboarding of new models without full-scale retraining. ZeroRouter features a context-aware predictor that maps queries to this universal space and a dual-mode optimizer that balances accuracy, cost, and latency. Our framework consistently outperforms all baselines, delivering higher accuracy at lower cost and latency.

cs.LG↗

arXiv2026

FalconFS: Distributed File System for Large-Scale Deep Learning Pipeline

Client-side metadata caching has long been considered an effective method for accelerating metadata operations in distributed file systems (DFSs). However, we have found that client-side state (e.g., caching) is not only ineffective but also consumes valuable memory resources in the deep learning pipelines. We thus propose FalconFS, a DFS optimized for deep learning pipelines with the stateless-client architecture. Specifically, instead of performing client-side path resolution and caching, FalconFS efficiently resolves paths on the server side using hybrid metadata indexing and lazy namespace replication. FalconFS also boosts server concurrency with concurrent request merging and provides easy deployment with VFS shortcut. Evaluations against CephFS and Lustre show that FalconFS achieves up to 5.72$\times$ throughput for small file read/write and up to 12.81$\times$ throughput for deep learning model training. FalconFS has been running in Huawei autonomous driving system's production environment with 10,000 NPUs for one year and has been open-sourced.

cs.DC↗