arXiv subjects

Ning Lu

Publications and source records attributed to Ning Lu.

At least 19 recordsLinked to original sources

arXiv2026

RoboPhys-3D: A Comprehensive Embodied World Model Evaluation via 3D Reconstruction

Video world models increasingly serve as data engines, action planners, and simulators for embodied AI, but conventional embodied world model (EWM) benchmarks lack a unified 3D-grounded protocol for establishing whether generated rollouts preserve the underlying 3D scene state or translate into executable actions. We introduce RoboPhys-3D, a 3D-grounded EWM benchmark built on RoboTwin 2.0, covering 50 manipulation tasks across four regimes, with 5,000 episodes and 25,000 multi-view ground-truth videos. A defining feature of RoboPhys-3D is that generated and ground-truth videos are processed through the same 3D reconstruction pipeline, enabling reconstruction-induced error to be distinguished from generation-induced error. The RoboPhys-3D benchmark organizes 50 complementary metrics into 18 sub-dimensions across four levels: pixel-level fidelity, 3D geometry consistency, state-level understanding, and task-level completeness. We further introduce Average Full Score, a hierarchical score averaging all 50 metrics for comprehensive evaluation, and RoboPhyscore, a compact task-aligned score averaging the metrics most strongly correlated with task success. Among the four representative video world models, Cosmos 3 achieves the highest RoboPhyscore (0.6330, 92.7% of ground truth), while state- and execution-grounded metrics reveal substantial failures that perceptual and vision-language model-based judgments fail to capture. RoboPhyscore further exhibits strong agreement with human evaluation (Pearson r = 0.9761 and Spearman \r{ho} = 0.8962), demonstrating the importance of grounded, execution-aware evaluation for EWM capability.

cs.RO↗

arXiv2026

MuEvo: LLM-Driven Evolution of Multi-Heuristic Ensemble

Large language model-based automated heuristic design (LLM-AHD) has shown strong potential in discovering effective heuristics for combinatorial optimization problems. However, existing methods primarily optimize a single heuristic, whereas practical optimization frameworks often rely on multiple interacting components. Directly extending single-heuristic methods is challenging because early component selection can overlook components with late potential, while independent evolution ignores inter-component dependencies. We propose MuEvo, an LLM-driven framework for evolving heuristic ensembles under ensemble-level feedback. MuEvo combines Dynamic Component Management, which uses short-budget probing and a reversible lifecycle to revise component priorities throughout the search, with LLM-Driven Co-Evolution, which coordinates component populations through Multi-Ensemble Evaluation, Cross-Component Information Sharing, Relation-Guided Pair Evolution, and Adaptive Budget Allocation. We evaluate MuEvo on selection hyper-heuristics and componentized ant colony optimization across four combinatorial optimization domains. Results show that MuEvo consistently improves human-designed frameworks and outperforms representative multi-component extensions of state-of-the-art LLM-AHD methods, demonstrating its effectiveness across both controller-mediated heuristic pools and functionally differentiated algorithmic components.

cs.NE↗

arXiv2026

Cooperative Coevolution for Resource-Constrained Agentic LLM Post-Training

Tool-using large language model (LLM) agents produce long, multi-turn trajectories, making gradient-based post-training memory-intensive. Evolution strategies (ES) enable memory-efficient full-parameter post-training without backpropagation and can eventually match the performance of gradient-based reinforcement learning (RL). However, resource-constrained settings typically offer only a few GPUs, so the high GPU-hour requirements of ES translate into prohibitively long training times. To address this, we introduce Cooperative Parameter-subspace Evolution Strategy (CoPES), a cooperative coevolutionary method that decomposes the full parameter space into lower-dimensional subspaces and searches over them cooperatively to improve optimization efficiency. We post-train a Qwen3.5-4B tool-using agent for the math task and evaluate it on five benchmarks of varying difficulty. Under the GPU-hour budget of full-parameter GRPO's best validation checkpoint, CoPES recovers 92% of GRPO's validation-accuracy gain, versus 67% for standard ES, while its theoretical GPU memory requirement is less than one-eighth that of full-parameter GRPO. It consistently outperforms standard ES and LoRA-based GRPO on all evaluated pass@k metrics across the five benchmarks. Additional experiments further show the advantage of CoPES on the question-answering task. These results demonstrate an improved trade-off between memory requirements and training time for agentic LLM post-training under resource constraints. The code is open-sourced in https://github.com/MetaronWang/CoPES

cs.AI↗

arXiv2026

VizRAG: Enhancing Retrieval-Augmented Generation with Hypergraph Visualization

Hypergraph-based RAG systems surpass traditional graph-based approaches by organizing complex n-ary atomic facts among entities, rather than relying solely on binary relationships. Despite the advancements in multimodal large language models (MLLMs) with enhanced visual capabilities, current hypergraph-based RAG frameworks predominantly restrict knowledge retrieval and reconstruction to a unimodal, text-centric paradigm. This limitation prevents them from fully leveraging the powerful visual perception capabilities of modern MLLMs. To address this gap, we systematically explore the integration of hypergraph awareness in RAG systems through visual cues. By incorporating visual representations of hypergraphs into the RAG pipeline, we introduce VizRAG, the first RAG system to support visual hypergraph structure awareness. Experimental results demonstrate that VizRAG significantly outperforms strong baselines, validating the promising potential of hypergraph visualization as a novel approach for RAG systems.

cs.CL↗

arXiv2026

Policy and World Modeling Co-Training for Language Agents

Reinforcement learning (RL) improves large language model (LLM) agents by teaching them which actions lead to high rewards, but provides little supervision on what those actions do to the environment. World modeling (WM) can fill this gap, yet existing approaches often require separate simulators, extra training stages, or additional inference-time computation. We observe that on-policy RL rollouts already contain the needed signal: each transition pairs an action with its resulting next observation. Based on this observation, we propose PaW, a Policy and World modeling co-training framework that adds auxiliary WM supervision to the same policy during RL, without changing the inference paradigm. To make auxiliary WM supervision informative and stable, PaW introduces three components: action-entropy-based WM data selection, noise-tolerant WM loss, and reward-adaptive loss balancing. Experiments on three agentic task benchmarks show consistent improvements over strong RL baselines across models and RL algorithms. These results suggest that standard RL rollouts are a practical source of WM supervision for language-agent training.

cs.LG↗

arXiv2026

AHD Agent: Agentic Reinforcement Learning for Automatic Heuristic Design

Automatic heuristic design (AHD) has emerged as a promising paradigm for solving NP-hard combinatorial optimization problems (COPs). Recent works show that large language models (LLMs), when integrated into well-designed frameworks (i.e., LLM-AHD), can autonomously discover high-performing heuristics. However, existing LLM-AHD frameworks typically treat LLMs as passive generators within fixed workflows, where the model generates heuristics from manually designed, limited context. Such context may fail to capture state-dependent information (e.g., specific failure modes), leading to inefficient trial-and-error exploration. To overcome these limitations, we propose AHD Agent, a novel tool-integrated, multi-turn framework that empowers LLMs to proactively decide whether to generate heuristics or invoke tools to retrieve targeted evidence from the solving environment. To effectively train such a dynamic decision-making agent, we introduce an agentic reinforcement learning (RL) system, which leverages a novel environment synthesis pipeline to optimize a compact model's generalizable AHD capabilities. Experiments across eight diverse domains, including four held-out tasks, demonstrate that our 4B-parameter agent matches or surpasses state-of-the-art baselines using much larger models, while requiring significantly fewer evaluations. Model and inference scaling analysis further reveals that AHD Agent offers an effective trajectory toward truly autonomous heuristic design.

cs.AI↗

arXiv2026

Accurate Frequency Response Modeling in Integrated T&D Co-Simulation via EWMA-RTTA-Based Quadratic Extrapolation

The large-scale integration of inverter-based resources (IBRs), particularly distributed photovoltaics (DPVs), into distribution networks increases the need for integrated transmission and distribution (T&D) co-simulation. A key challenge in such co-simulation lies in accurately modeling system frequency across two asynchronous simulation environments. For example, the transmission system, simulated in the phasor domain, can operate with a simulation timestep of 10 ms, while the distribution system, simulated in the electromagnetic transient domain (EMT) to include IBR models, uses a much finer timestep of 100 microseconds. To ensure accurate PLL-based frequency estimation in distribution systems, it is essential to predict voltage magnitude and phase angle variations within the 10 ms transmission intervals, rather than using constant values that cause inaccurate frequency calculations. This issue becomes particularly critical when modeling primary and secondary frequency response services provided by IBRs. To address this challenge, we propose an automated Exponentially Weighted Moving Average Real-Time Threshold Adaptation (EWMA-RTTA) method, which utilizes Quadratic Extrapolation to predict voltage magnitude and phase angle trends more precisely. The proposed method is validated using two Opal-RT simulators: one simulating an IEEE 118-bus transmission system and the other simulating an IEEE 123-bus distribution network. Simulation results demonstrate that our approach improves the normalized mean absolute error (nMAE) by a factor of 25.7 compared to methods that do not account for time mismatches, offering a scalable and accurate solution for modeling IBR-based frequency response in modern power systems.

eess.SY↗

arXiv2026

Train at Moving Edge: Online-Verified Prompt Selection for Efficient RL Training of Large Reasoning Model

Reinforcement learning (RL) has become essential for post-training large language models (LLMs) in reasoning tasks. While scaling rollouts can stabilize training and enhance performance, the computational overhead is a critical issue. In algorithms like GRPO, multiple rollouts per prompt incur prohibitive costs, as a large portion of prompts provide negligible gradients and are thus of low utility. To address this problem, we investigate how to select high-utility prompts before the rollout phase. Our experimental analysis reveals that sample utility is non-uniform and evolving: the strongest learning signals concentrate at the ``learning edge", the intersection of intermediate difficulty and high uncertainty, which shifts as training proceeds. Motivated by this, we propose HIVE (History-Informed and online-VErified prompt selection), a dual-stage framework for data-efficient RL. HIVE utilizes historical reward trajectories for coarse selection and employs prompt entropy as a real-time proxy to prune instances with stale utility. By evaluating HIVE across multiple math reasoning benchmarks and models, we show that HIVE yields significant rollout efficiency without compromising performance.

cs.LG↗

arXiv2026

Learning Domain Agnostic Latent Embeddings of 3D Faces for Zero-shot Animal Expression Transfer

We present a zero-shot framework for transferring human facial expressions to 3D animal face meshes. Our method combines intrinsic geometric descriptors (HKS/WKS) with a mesh-agnostic latent embedding that disentangles facial identity and expression. The ID latent space captures species-independent facial structure, while the expression latent space encodes deformation patterns that generalize across humans and animals. Trained only with human expression pairs, the model learns the embeddings, decoupling, and recoupling of cross-identity expressions, enabling expression transfer without requiring animal expression data. To enforce geometric consistency, we employ Jacobian loss together with vertex-position and Laplacian losses. Experiments show that our approach achieves plausible cross-species expression transfer, effectively narrowing the geometric gap between human and animal facial shapes.

cs.CV↗

arXiv2025

VL-RouterBench: A Benchmark for Vision-Language Model Routing

Multi-model routing has evolved from an engineering technique into essential infrastructure, yet existing work lacks a systematic, reproducible benchmark for evaluating vision-language models (VLMs). We present VL-RouterBench to assess the overall capability of VLM routing systems systematically. The benchmark is grounded in raw inference and scoring logs from VLMs and constructs quality and cost matrices over sample-model pairs. In scale, VL-RouterBench covers 14 datasets across 3 task groups, totaling 30,540 samples, and includes 15 open-source models and 2 API models, yielding 519,180 sample-model pairs and a total input-output token volume of 34,494,977. The evaluation protocol jointly measures average accuracy, average cost, and throughput, and builds a ranking score from the harmonic mean of normalized cost and accuracy to enable comparison across router configurations and cost budgets. On this benchmark, we evaluate 10 routing methods and baselines and observe a significant routability gain, while the best current routers still show a clear gap to the ideal Oracle, indicating considerable room for improvement in router architecture through finer visual cues and modeling of textual structure. We will open-source the complete data construction and evaluation toolchain to promote comparability, reproducibility, and practical deployment in multimodal routing research.

cs.LG↗

arXiv2025

A voltage-responsive strongly dipolar-coupled macrospin network with emergent dynamics for computing

Emergent behavior, which arises from local interactions between simple elements, is pervasive in nature. It underlies the energy-efficient computing in our brains. However, realizing such dynamics in artificial materials, particularly under low-energy stimuli, remains a fundamental challenge. While dipole-dipole interactions are typically suppressed in magnetic storage, here we harness and amplify them to construct a strongly dipolar-coupled network of SmCo5 macrospins at wafer scale, which can exhibit intrinsic interaction-driven collective dynamics in response to voltage pulses. The network combines three essential ingredients: strong dipolar coupling by large single-domain macrospin, giant voltage control of coercivity over nearly 1000-fold, and disordered network topology with frustrated Ising-like energy landscape. When stimulated by 1 V pulses, the network enters a regime where interaction-driven magnetic behaviors emerge, including spontaneous demagnetization, greatly enhanced magnetization modulation, reversible freeze and resume evolution and stochastic convergence toward low-energy magnetic configurations. All these behaviors are completely absent at the single-nanomagnet level. Furthermore, by constructing micromagnetic models of the strongly dipolar-coupled macrospin networks, we show that the resulting nonlinear, high-dimensional collective dynamics, intrinsic to strongly-interacting systems, can enable accurate chaotic Mackey-Glass prediction and multiclass drone-signal classification. Our work establishes the voltage-responsive strongly-coupled SmCo5 network as a mesoscopic platform for probing emergent magnetic dynamics previously inaccessible under ambient conditions. It also suggests a fundamental distinct route towards scalable, low-voltage computing, one rooted in native physical interaction-driven collective dynamics at the network level.

cond-mat.mtrl-sci↗

arXiv2025

The Regulated GeAs Cycles with the New $^{63}$Ga(p,$\gamma$)$^{64}$Ge and $^{64}$Ge(p,$\gamma$)$^{65}$As Reaction Rates and Their Impact on the GS 1826$-$24 Clocked Bursts and SAX J1808.4$-$3658 Photospheric Radius Expansion Bursts

The $^{63}$Ga(p,$\gamma$)$^{64}$Ge and $^{64}$Ge(p,$\gamma$)$^{65}$As thermonuclear reactions connect the ZnGa and GeAs cycles by diverting the flow of the rapid proton capture process from $^{63}$Ga to $^{65}$As. Changes in these two reaction rates regulate the ZnGa and GeAs cycles and may affect the modeled properties matching with the observed counterparts of a type I X-ray burster. We implement the latest $^{63}$Ga(p,$\gamma$)$^{64}$Ge and $^{64}$Ge(p,$\gamma$)$^{65}$As reaction rates to the state-of-the-art self-consistent one-dimensional multi-zone thermo-hydrodynamic code, KEPLER, to study the influence of these new reaction rates on the models of the GS 1826$-$24 clocked burster and SAX J1808.4$-$3658 photospheric radius expansion burster. Both new reaction rates obtained by Lu et al. [Phys. Rev. C 110, 065804 (2024)] are determined from complementing the experimental input with the nuclear spectroscopic information deduced from the full pf-shell space configuration-interaction shell-model calculations. By constraining the models on reproducing the observed burst peak, light-curve profile, fluence, and recurrence time, we find that the impact of the newly measured proton thresholds and respective proton-capture reactions on the burst light-curve profile of the GS 1826$-$24 clocked burster is, in fact, not as significant as claimed by Zhou et al. [Nat. Phys. 19, 1091 (2023)]. With or without the inclusion of the newly determined reaction rate of the highly influential $^{22}$Mg($\alpha$,p)$^{25}$Al reaction, the impact of the new $^{63}$Ga(p,$\gamma$)$^{64}$Ge and $^{64}$Ge(p,$\gamma$)$^{65}$As reaction rates on SAX J1808.4$-$3658 photospheric radius expansion bursts is evident. Our finding indicates that the models reproducing the 2002 October epoch of SAX J1808.4$-$3658 photospheric radius expansion burster is more sensitive to the uncertainties of thermonuclear reaction rates.

astro-ph.HE↗

arXiv2025

An eco-friendly universal strategy via ribavirin to achieve highly efficient and stable perovskite solar cells

The grain boundaries of perovskite films prepared by the solution method are highly disordered, with a large number of defects existing at the grain boundaries. These defect sites promote the decomposition of perovskite. Here, we use ribavirin obtained through bacillus subtilis fermentation to regulate the crystal growth of perovskite, inducing changes in the work function and energy level structure of perovskite, which significantly reduces the defect density. Based on density functional theory calculations, the defect formation energies of VI, VMA, VPb, and PbI in perovskite are improved. This increases the open-circuit voltage of perovskite solar cells (PSCs) (ITO/PEDOT:PSS/perovskite/PCBM/BCP/Ag) from 1.077 to 1.151 V, and the PCE increases significantly from 17.05% to 19.86%. Unencapsulated PSCs were stored in the environment (humidity approximately 35+-5%) for long-term stability testing. After approximately 900 hours of storage, the PCE of the ribavirin-based device retains 84.33% of its initial PCE, while the control-based device retains only 13.44% of its initial PCE. The PCE of PSCs (ITO/SnO2/perovskite/Spiro-OMETAD/Ag) is increased from 20.16% to 22.14%, demonstrating the universality of this doping method. This universal doping strategy provides a new approach for improving the efficiency and stability of PSCs using green molecular doping strategies.

cond-mat.mtrl-sci↗

arXiv2025

LLM-Driven Instance-Specific Heuristic Generation and Selection

Combinatorial optimization problems are widely encountered in real-world applications. A critical research challenge lies in designing high-quality heuristic algorithms that efficiently approximate optimal solutions within a reasonable time. In recent years, many works have explored integrating Large Language Models (LLMs) with Evolutionary Algorithms to automate heuristic algorithm design through prompt engineering. However, these approaches generally adopt a problem-specific paradigm, applying a single algorithm across all problem instances, failing to account for the heterogeneity across instances. In this paper, we propose InstSpecHH, a novel framework that introduces the concept of instance-specific heuristic generation. InstSpecHH partitions the overall problem class into sub-classes based on instance features and performs differentiated, automated heuristic design for each problem subclass. By tailoring heuristics to the unique features of different sub-classes, InstSpecHH achieves better performance at the problem class level while avoiding redundant heuristic generation for similar instances, thus reducing computational overhead. This approach effectively balances the trade-off between the cost of automatic heuristic design and the quality of the obtained solutions. To evaluate the performance of InstSpecHH, we conduct comprehensive experiments on 4,500 subclasses of the Online Bin Packing Problem (OBPP) and 365 subclasses of the Capacitated Vehicle Routing Problem (CVRP). Experimental results show that InstSpecHH demonstrates strong intra-subclass and inter-subclass generalization capabilities. Compared to previous problem-specific methods, InstSpecHH reduces the average optimality gap by 6.06\% for OBPP and 0.66\% for CVRP. These results highlight the potential of instance-aware automatic heuristic design to further enhance solution quality.

cs.NE↗

arXiv2025

Safe Delta: Consistently Preserving Safety when Fine-Tuning LLMs on Diverse Datasets

Large language models (LLMs) have shown great potential as general-purpose AI assistants across various domains. To fully leverage this potential in specific applications, many companies provide fine-tuning API services, enabling users to upload their own data for LLM customization. However, fine-tuning services introduce a new safety threat: user-uploaded data, whether harmful or benign, can break the model's alignment, leading to unsafe outputs. Moreover, existing defense methods struggle to address the diversity of fine-tuning datasets (e.g., varying sizes, tasks), often sacrificing utility for safety or vice versa. To address this issue, we propose Safe Delta, a safety-aware post-training defense method that adjusts the delta parameters (i.e., the parameter change before and after fine-tuning). Specifically, Safe Delta estimates the safety degradation, selects delta parameters to maximize utility while limiting overall safety loss, and applies a safety compensation vector to mitigate residual safety loss. Through extensive experiments on four diverse datasets with varying settings, our approach consistently preserves safety while ensuring that the utility gain from benign datasets remains unaffected.

cs.LG↗

arXiv2025

Is PRM Necessary? Problem-Solving RL Implicitly Induces PRM Capability in LLMs

The development of reasoning capabilities represents a critical frontier in large language models (LLMs) research, where reinforcement learning (RL) and process reward models (PRMs) have emerged as predominant methodological frameworks. Contrary to conventional wisdom, empirical evidence from DeepSeek-R1 demonstrates that pure RL training focused on mathematical problem-solving can progressively enhance reasoning abilities without PRM integration, challenging the perceived necessity of process supervision. In this study, we conduct a systematic investigation of the relationship between RL training and PRM capabilities. Our findings demonstrate that problem-solving proficiency and process supervision capabilities represent complementary dimensions of reasoning that co-evolve synergistically during pure RL training. In particular, current PRMs underperform simple baselines like majority voting when applied to state-of-the-art models such as DeepSeek-R1 and QwQ-32B. To address this limitation, we propose Self-PRM, an introspective framework in which models autonomously evaluate and rerank their generated solutions through self-reward mechanisms. Although Self-PRM consistently improves the accuracy of the benchmark (particularly with larger sample sizes), analysis exposes persistent challenges: The approach exhibits low precision (<10\%) on difficult problems, frequently misclassifying flawed solutions as valid. These analyses underscore the need for continued RL scaling to improve reward alignment and introspective accuracy. Overall, our findings suggest that PRM may not be essential for enhancing complex reasoning, as pure RL not only improves problem-solving skills but also inherently fosters robust PRM capabilities. We hope these findings provide actionable insights for building more reliable and self-aware complex reasoning models.

cs.AI↗

arXiv2025

SemDiff: Generating Natural Unrestricted Adversarial Examples via Semantic Attributes Optimization in Diffusion Models

Unrestricted adversarial examples (UAEs), allow the attacker to create non-constrained adversarial examples without given clean samples, posing a severe threat to the safety of deep learning models. Recent works utilize diffusion models to generate UAEs. However, these UAEs often lack naturalness and imperceptibility due to simply optimizing in intermediate latent noises. In light of this, we propose SemDiff, a novel unrestricted adversarial attack that explores the semantic latent space of diffusion models for meaningful attributes, and devises a multi-attributes optimization approach to ensure attack success while maintaining the naturalness and imperceptibility of generated UAEs. We perform extensive experiments on four tasks on three high-resolution datasets, including CelebA-HQ, AFHQ and ImageNet. The results demonstrate that SemDiff outperforms state-of-the-art methods in terms of attack success rate and imperceptibility. The generated UAEs are natural and exhibit semantically meaningful changes, in accord with the attributes' weights. In addition, SemDiff is found capable of evading different defenses, which further validates its effectiveness and threatening.

cs.LG↗

arXiv2025

Design and Implementation of Scalable Communication Interfaces for Reliable and Stable Real-time Co-Simulation of Power Systems

Co-simulation offers an integrated approach for modeling the large-scale integration of inverter-based resources (IBRs) into transmission and distribution grids. This paper presents a scalable communication interface design and implementation to enable reliable and stable real-time co-simulation of power systems with high IBR penetration. The communication interface is categorized into two types: local and remote. In local scenarios, where subsystems are connected within a single local area network (LAN), low-latency communication facilitates the seamless integration of electromagnetic transient (EMT) and phasor-domain models, enabling efficient interactions with power and energy management algorithms. For remote scenarios, data exchange is achieved via internet-based file sharing or VPN-enabled communication. The performance of both methods is evaluated using OPAL-RT as a real-time simulator, demonstrating scalability, effectiveness, and challenges specific to real-time co-simulation applications. To mitigate instability arising from data resolution mismatches in time-sensitive co-simulations, a real-time data extrapolation method is proposed. This approach significantly enhances stability and reliability, ensuring more accurate simulation outcomes. The implementation code is available on GitHub, providing researchers the tools to replicate and expand upon this work.

eess.SY↗