In response to the increasing demand for heterogeneous data interaction and cross-disciplinary modeling in aerospace situational awareness simulations, we propose a microservice platform for data-model computing (MP-DMC), which is a high-performance microservice framework built on a container cloud. The proposed MP-DMC framework unifies the control of data and models through a distributed node resource management and scheduling strategy, integrating an election-optimized leader-follower mechanism, a predictive model based on a double-moving-average long short-term memory (DMA-LSTM) network for dynamic elastic scaling, and an intelligent load migration algorithm to address management inefficiencies, prevent node crashes, and mitigate resource oscillations under high-concurrency conditions. Experimental results demonstrate that the proposed MP-DMC framework outperforms mainstream algorithms in terms of election performance, node scaling efficiency, task response time, and load balancing, including consensus algorithms (Paxos, Raft, and PBFT), elastic scaling methods (HPA, DMA-HPA, ProSmart HPA, and RL), and scheduling algorithms (round-robin, purely random, and weighted random), achieving exceptional resource allocation performance and system availability.
Conventional circuit systems typically exhibit limited capability in suppressing disturbances with strong temporal structures. This study proposes a bio-inspired anti-interference framework based on neural synchronization mechanisms. Hodgkin-Huxley (HH) and Izhikevich (IZH) networks with different topologies are compared in terms of synchronization stability, energy deviation, and hardware feasibility under external interference. The proposed framework integrates the neuron models and topological structures of neural networks to establish a unified evaluation system for bio-inspired anti-interference circuits. The simulation results demonstrate that the HH network exhibits stronger robustness under strong coupling, maintaining a synchronization index near 1.0 and a lower relative energy deviation than the IZH network. The small-world topology achieves the best disturbance tolerance, whereas the chain topology exhibits the weakest robustness for both models. In addition, hardware implementation demonstrates that the IZH network significantly reduces lookup table consumption, and both the HH and IZH circuits maintain rhythmic firing and waveform integrity under 20 mV square-wave interference. These results indicate that neural synchronization mechanisms provide an effective and practically implementable mechanism for adaptive anti-interference circuit design.
Large language models (LLMs) have achieved strong performance on many language tasks, but they still struggle with culturally grounded symbolic reasoning. Existing benchmarks have not systematically evaluated the progressive capability chain required for cross-cultural understanding, which involves intra-cultural symbolic understanding, cross-cultural symbolic alignment, and cross-cultural conflict identification. To address this gap, we propose CCU-Bench, a systematic benchmark for evaluating cross-cultural understanding in LLMs. Grounded in Hofstede's cultural dimensions theory, CCU-Bench focuses on three culturally distinct contexts, China, Japan, and Mexico, and is organized around three dimensions, image, connotation, and emotion. The benchmark is constructed through three stages, data collection, question generation, and quality control, resulting in a high-quality evaluation set of 3029 question-answer pairs in five formats. Experiments on 19 mainstream LLMs show that current models perform unsatisfactorily on this task, achieving an average score of 57.8%. Closed-source models consistently outperform open-source models, while cross-cultural symbolic alignment remains the most challenging sub-task. Further error analysis reveals that the dominant failures stem from deficiencies in cultural knowledge, biases in intent mapping, and weak higher-order reasoning about cross-cultural conflicts, rather than simple instruction-following issues. These findings highlight persistent limitations of current LLMs in culturally grounded reasoning and demonstrate that CCU-Bench provides a standardized benchmark for culturally aware artificial intelligence research.
Graphical user interface(GUI)agents are widely used in general-purpose digital automation.Large language models,vision-language models,and multimodal foundation models can now interpret screenshots,follow natural-language instructions,and execute grounded actions in websites,mobile applications,and desktop operating systems.However,despite their different execution conditions,these settings are commonly grouped under one label.This survey reviews representative frameworks,models,datasets,and benchmarks of GUI agents through a common technical stack:observation,grounding,planning,memory,execution,and verification.We advocate the treatment of web,mobile,and desktop/operating-system agents as distinct operational regimes with different observabilities,action semantics,hidden-state dependences,execution costs,and side-effect risks.We also describe a shift from next-action prediction to dependable execution with increased focus on outcome verification,execution efficiency,and trustworthiness.The survey concludes with future directions in belief-state tracking,adaptive sensing,hybrid GUI-tool execution,and human oversight for reliable computer use.
This paper introduces a novel knee exoskeleton designed to provide assistive torque to the knee joint and facilitate postoperative rehabilitation.Drawing inspiration from the anatomy of the human knee joint,the exoskeleton incorporates a biomimetic structure that integrates the"optimal axis of rotation"with the trajectory of the instantaneous center of rotation.This design effectively mitigates the issue of kinematic misalignment in the human‒machine joint,offering a more comfortable wearing experience for users.Furthermore,a highly integrated quasi-direct drive system is incorporated into the knee exoskeleton.The drive system can deliver a continuous torque of 18 N·m and a peak torque of 45 N·m to sufficiently meet the daily walking needs of patients.From a control perspective,a unified sliding mode control strategy based on iterative learning control is proposed to maintain high tracking accuracy.This strategy enables the system to achieve accurate tracking of the knee motion trajectory by continuously updating the control inputs through successive gait cycles.Experimental results demonstrate that the designed biomimetic joint mechanism effectively resolves the issue of misalignment in the human‒machine joint.Furthermore,the synergistic control strategy demonstrates excellent steady-state precision and transient robustness in terms of trajectory tracking during human‒exoskeleton cooperative walking.
Three-dimensional convolutional neural networks (3D CNNs) show considerable promise for lung nodule detection. However, their high computational complexity and memory demands present substantial challenges for acceleration on a single field-programmable gate array (FPGA). To address this, we propose efficient mapping schemes for a multi-FPGA platform, leveraging its massive parallelism to maximize computational efficiency. Our system, integrating six customized FPGA boards, achieves state-of-the-art performance, delivering approximately 15.9 tera operations per second (TOPS) for nodule segmentation and approximately 3.8 TOPS for nodule classification. Compared to a central processing unit baseline, it achieves a 128.2× speedup while exhibiting 6.7× higher energy efficiency than a graphics processing unit implementation. Furthermore, the system attains a state-of-the-art recall rate of 87.1% on the real-world clinical benchmark.
Open multi-agent systems (OMASs), characterized by the dynamic joining and leaving of agents, possess distinct attributes such as agent-level autonomy, time-varying network topologies, and environmental openness. These characteristics make them highly applicable to dynamic scenarios like robotic swarms, smart grids, and vehicular networks. However, such dynamism introduces core challenges in maintaining system stability, achieving efficient collaboration, and guaranteeing decision robustness. This paper presents a brief overview of recent advances in distributed control and decision-making algorithms for OMASs. First, the fundamental concepts and control strategies of OMASs are systematically reviewed. Second, distributed decision-making mechanisms encompassing distributed consensus optimization, separable resource allocation, and Nash equilibrium (NE) seeking in non-cooperative games are discussed, highlighting key technologies and typical methods. Finally, an outlook on future perspectives in the field is presented.
Chinese medical named entity recognition (CMNER) is a fundamental task in medical information extraction. It is crucial for building downstream applications, such as clinical knowledge graphs, and enabling intelligent clinical decision-making. However, existing mainstream approaches, including lexicon-enhanced, span-based, and grid-based tagging methods, struggle with the absence of natural boundaries, complex nested entity structures, and long-range contextual dependencies inherent in clinical texts. To address these challenges, we propose a novel dual-stream fusion with dynamic grid interaction (D2GI) model, which performs deep semantic mining by integrating complementary feature streams and adaptive grid interactions to accurately capture complex entity boundaries and inter-character relations. Specifically, our dual-stream fusion architecture leverages RoFormer to extract long-range dependencies and incorporates Word2Vec to provide stable prior semantics, thereby enhancing the representation of rare medical terms. Furthermore, to overcome the limitations of static refinement, the dynamic grid interaction module employs a gated attention mechanism to adaptively fuse local and global contexts, facilitating accurate recognition of nested entities. Multiple experiments on three public datasets demonstrate that D2GI is superior to state-of-the-art baselines, achieving F1-score improvements of 1.48 percentage points (PPs) on CMeEE-V2, 1.61 PPs on DiaKG, and 3.08 PPs on CCKS2020.
This study proposes a fast zeroing neural network algorithm to solve time-varying Laplacian linear systems arising from the modeling of superstructure quadrilateral dynamic resistor networks on hammock surfaces. By incorporating the intrinsic structure of the underlying special-form matrices into the core neurodynamic design, the proposed algorithm enables efficient real-time computation of electric potentials under dynamic conditions. The Lyapunov-based analysis proves global exponential convergence. Numerical simulations on resistor networks of various scales demonstrate high computational efficiency and verify convergence to solutions from arbitrary initial conditions. Furthermore, by integrating the proposed algorithm as a potential field solver with a directional potential field path planning algorithm and exploiting the natural descent property of resistor network node potentials, we propose a fast path planning algorithm for robotic navigation on hammock surfaces. Compared with conventional path planning approaches, the proposed algorithm achieves higher computational efficiency in the aforementioned hammock surface path planning task, and this advantage becomes increasingly pronounced as the scale increases. The proposed algorithm is also applied to dynamic path planning tasks, further validating its potential in robotics and control applications. Finally, we present two conjectures.
This study proposes an efficiency-enhanced dual-mode orthogonal Doherty power amplifier(ODPA)architecture.The operation theory,design process,and implementation of the proposed ODPA are introduced in detail.By introducing an orthogonal architecture with control signal power injection and incorporating a modified output matching network,the efficiency of Doherty power amplifiers under active load modulation can be enhanced.Furthermore,owing to the additional degrees of freedom provided by the orthogonal architecture,which enables independent amplitude and phase control and dynamic reshaping of the load modulation trajectories,and the dual-mode reciprocal bias configuration,the proposed ODPA exhibits high adaptability to load mismatch.To validate the proposed architecture,a 2.1-GHz dual-mode ODPA is designed and fabricated.With a matched load,the fabricated dual-mode ODPA exhibites over 60%drain efficiency throughout the power range from 6-dB output back-off(OBO)to saturation.For mismatched loads with a 2∶1 voltage standing wave ratio,a saturated output power of 42.8-44.6 dBm and OBO efficiency of 47.2%-58.1%are obtained through mode reconfiguration.
This paper addresses the problem of end-effector position-tracking control for micro-nano free-floating space robots in Cartesian space without relying on explicit analytical kinematic or dynamic models. To address this challenge, we develop a two-layer learning architecture. In the first layer, a deep neural network is used for kinematic learning to capture the nonlinear mapping from end-effector Cartesian coordinates to joint angular velocities and to generate reference joint trajectories. In the second layer, a Koopman-operator-based network is employed to construct an approximately linearized representation of the joint-space dynamics of free-floating space robots. Based on this model, we propose a terminal fractional-order model predictive control scheme that incorporates the Grünwald-Letnikov fractional-order operator, thereby enhancing online control performance and improving tracking speed and accuracy relative to conventional model predictive control. Simulation results verify the effectiveness of the proposed method, demonstrating accurate and rapid end-effector trajectory tracking, all without requiring explicit analytical kinematic and dynamic models in the controller design, while the training pipeline relies solely on input-output trajectories.
The dual transponder carrier ranging system, an enhancement over the dual one-way ranging system, provides high precision relative distance measurements without imposing stringent time synchronization requirements. However, a critical limitation to its use in absolute distance measurement is the initial value change phenomenon, which manifests as abrupt phase changes after every system restart. Through theoretical analysis, the root causes of this phenomenon are identified as a fractional output-to-input frequency ratio in the frequency synthesizer and a non-integer coherent turnaround ratio. To validate the theoretical analysis, simulations are conducted using Simulink, and experimental verification is performed on an S-band and C-band hardware platform. Both simulation and experimental results demonstrate that configuring both ratios as integers effectively suppresses these abrupt phase changes. These findings establish an integer-ratio design rule that is independent of specific frequency plans, providing a general design guideline for coherent transponder ranging systems intended for absolute distance measurement, as well as a reference for frequency planning in other absolute distance measurement systems of the same kind.
In absolute distance measurement and positioning applications, atmospheric refraction error is a critical factor limiting measurement accuracy. Temperature plays a dominant role in computing the atmospheric refractive index. However, accurately acquiring the temperature field along the ranging path in complex and dynamic outdoor environments remains challenging due to limited sensor deployment and environmental nonstationarity. We propose a spatiotemporal temperature data fusion method for atmospheric refraction correction, which integrates the strengths of the generalized regression neural network (GRNN) and Kriging interpolation within a Kalman filter. This method achieves dynamic prediction and high-accuracy reconstruction of temperature parameters. The proposed method is systematically validated through simulation analysis as well as indoor and kilometer-scale outdoor experimental measurements. The simulation results demonstrate that Kalman filter expanded fusion (KFEF) outperforms the traditional interpolation method radial basis function (RBF) and the state-of-the-art spatiotemporal interpolation and prediction methods spatiotemporal Kriging (STK) and Gaussian process (GP), in terms of both reconstruction accuracy and stability of the temperature field. Specifically, KFEF achieves a 61.54% reduction in root mean square error (RMSE) compared with RBF and reductions of 34.21% and 32.43% relative to STK and GP, respectively. This indicates its practical value for long-distance high-precision ranging engineering applications. Furthermore, the proposed spatiotemporal data fusion framework is highly general and scalable. It can also be applied to other temperature field prediction and reconstruction problems.
User-level file systems are widely adopted in research and production environments due to their flexibility and reduced risk of kernel crashes.Filesystem in Userspace(FUSE)is a general-purpose framework for developing user-level file systems in Linux.Compared with library-based file systems,FUSE adopts a cooperative architecture between kernel modules and userspace libraries,ensuring metadata security and compliance with standard portable operating system interface(POSIX)semantics.However,this architecture introduces substantial context-switching overhead.Existing optimization approaches for high-performance computing environments often improve input/output(I/O)performance at the expense of FUSE's cross-environment compatibility and kernel-level security guarantees.To address this limitation,this study proposes SplitFUSE,an I/O-acceleration framework for user-level file systems that preserves high compatibility and strong security.SplitFUSE introduces a split architecture that decouples metadata and data-request processing.Specifically,the kernel maintains full metadata consistency,while the userspace retains only a minimal metadata subset required to validate and process data requests that bypass the kernel securely.Implemented as a self-contained mechanism,SplitFUSE preserves the same cross-environment compatibility as conventional FUSE.Experimental results demonstrate that,under I/O-intensive small-write workloads,SplitFUSE achieves up to 4-6 times higher write bandwidth than native FUSE and outperforms state-of-the-art alternatives.For common file system workloads,it delivers substantial performance improvements with minimal migration overhead.
Accurate assessment of human exposure to millimeter-wave (mmWave) electric fields (E-fields) has recently become critical for public health and safety. High-spatial-resolution E-field distribution is required for assessment of mmWave electromagnetic exposure according to the International Electrotechnical Commission (IEC) and the Institute of Electrical and Electronics Engineers (IEEE) (IEC/IEEE 63195-2 standard). This study proposes a generative adversarial network (GAN) integrated with field gradient loss, termed EFGraGAN, for superresolution reconstruction of mmWave E-fields. The incorporation of E-field gradient loss enables the network to learn both local field magnitudes and spatial structures, thereby enhancing the accuracy and fine structural details of reconstructed E-field maps. To improve generalization across antenna types, the training dataset is generated using plane-wave integral representation (PWIR) and randomized parametric incidence, simulating diverse field distributions. Combined with bilinear interpolation, the method achieves high-resolution reconstruction at 30 GHz and 60 GHz, meeting the requirements of the IEC/IEEE 63195-2 standard for exposure assessment. Numerical simulations show that EFGraGAN reconstructs E-field distributions in a skin phantom with a maximum mean relative error (MRE) of <9% up to 60 GHz in a 4×4 dipole array scenario, outperforming conventional interpolation and traditional GAN methods. The approach also demonstrates strong robustness to noise, enabling current measurement systems to achieve accurate and efficient evaluation of mmWave exposure.
Reconstructing high dynamic range (HDR) images from a single low dynamic range (LDR) input requires recovering missing information in highlight-clipped and shadow-distorted regions. Existing methods generally rely on sufficient ground truth HDR images as supervision signals or multi-exposure LDR sequences to improve quality, limiting their flexibility. To address this, we propose USME-HDR, a framework for single-image HDR reconstruction based on multi-exposure priors, where the HDR reconstruction stage is learned without ground-truth HDR supervision. Specifically, an exposure-adjustment network (EAN) is trained in a supervised manner to map a single LDR image to over/under-exposure pairs. Inspired by the Retinex theory, we further decompose the input into a light map and a light feature, which are fed into the EAN as auxiliary inputs for luminance-aware exposure generation. An exposure time ratio guidance mechanism is further introduced to improve luminance fidelity. Finally, the HDR image is synthesized by fusing the original LDR image with generated multi-exposure images, refined through self-supervised optimization. Experiments demonstrate that during the test phase, USME-HDR reconstructs visually compelling HDR images from only a single LDR input, without requiring real low- or high-exposure images.
Three-dimensional network-on-chips (3D NoCs) are increasingly used to improve scalability in multicore systems. Through-silicon via (TSV) is a critical technology for enabling vertical interconnects between NoC layers. However, TSV-based interlayer connections are highly prone to faults resulting from manufacturing defects, aging, or other sources, which compromise system reliability. To address these challenges, particularly in chiplet-based 3D NoCs, robust fault-tolerant mechanisms are crucial for maintaining operational integrity in the presence of TSV faults. We introduce a novel fault-tolerant architecture designed to ensure persistent communication reliability despite permanent vertical link failures, named HyRAS, a hybrid redundancy- and serialization-based method. Our approach is built on two synergistic mechanisms. First, a lightweight spatial redundancy-based scheme leverages shared TSV resources to mitigate the impact of isolated faults. Second, for more severe fault scenarios, an adaptive serialization-based strategy is employed to maintain connectivity by efficiently using the remaining functional links. The architecture is rigorously evaluated through functional simulations using both synthetic traffic patterns and realistic application workloads. Compared to contemporary fault-tolerant methods, HyRAS achieves up to 28.2% higher throughput under realistic workloads with significant defect clusters. These gains are achieved with only modest overhead, incurring a 14.53% increase in area and 8.87% increase in power consumption relative to the standard redundancy-based router.
Renewable generation and load uncertainty pose significant challenges to power system security, necessitating efficient approaches to characterizing high-dimensional security regions. To overcome the curse of dimensionality, uncertainty neglect, and undue conservatism in existing methods, this paper proposes an approach integrating distributionally robust optimization (DRO) and deep learning for security region characterization. First, to properly account for uncertainty while avoiding excessive conservatism, a DRO-based active search strategy is developed to identify critical boundary points, where diffusion-generated renewable scenarios and load-deviation samples constructed around typical demand profiles are jointly used to build a robust probabilistic ambiguity set. Subsequently, a Transformer-based model learns from these boundary points to reconstruct the full high-dimensional security region. The model's self-attention mechanism captures the global nonlinear dependencies among dimensions, enabling a precise and efficient boundary fit. Simulations on IEEE test systems confirm that the approach accurately characterizes high-dimensional security regions at a low computational cost, yielding a security region with strong robustness to renewable-load uncertainty. This work offers a new paradigm for security assessment and decision support in power systems under high uncertainty.
Raft is a foundational consensus protocol for distributed systems,architected to ensure state machine replication and data consistency across machine clusters.However,traditional Raft faces significant performance bottlenecks,particu-larly regarding suboptimal election efficiency and substantial consensus latency in large-scale deployments.To address these challenges,this study presents MH-Raft,an enhanced consensus variant designed for high efficiency and minimal latency.We propose a hierarchical node management and election framework to optimize network coordination.Specifically,a leader election methodology leveraging the multi-objective evolutionary algorithm based on decomposition(MOEA/D)is formulated to minimize election latency by evaluating multi-dimensional node attributes.To further refine the proposed hierarchical architecture,a rigorous tightness definition is devised for optimal mediator node selection,which is integrated into a hybrid clustering algorithm that adaptively partitions the network and optimizes the mapping between mediator nodes and follower nodes.Quantitative evaluations via comprehensive experiments demonstrate that MH-Raft significantly reduces overall election latency and lowers consensus latency by 14.87%-34.45%,while enhancing average throughput by 30.43%compared to the conventional Raft implementation.
The location where a robot grasps an object is closely related to the task type. For the same object, different user requirements may necessitate different grasping strategies. Visual affordance serves as a reliable source of prior knowledge for manipulation. Existing methods learn affordance from images or videos, but planar affordance lacks the spatial information required for 6-degree-of-freedom (6-DoF) manipulation. Furthermore, current approaches are limited to affordances associated with predefined categories and cannot directly infer affordances from user instructions. To address such limitations, we propose a novel task: instruction-driven three-dimensional (3D) object affordance segmentation. To support this research, we introduce an instruction-affordance dataset (IAD), a challenging dataset consisting of 7190 object instances across 20 common object categories, paired with 624 manipulation instructions that specify the corresponding affordances. To evaluate generalization to novel commands, our dataset includes both seen and unseen settings. Building on this, we design an instruction-driven 3D affordance segmentation (IDAS) network, which extracts point cloud features and integrates instruction features layer by layer. Given a user instruction, our method segments suggested manipulation regions on the object's point cloud, thereby guiding the selection of optimal grasp poses. Experimental results show that our method outperforms other related approaches under both seen and unseen settings, demonstrating generalization ability to diverse user commands and unknown affordances.
