In Wireless Sensor Networks(WSNs),survivability is a crucial issue that is greatly impacted by energy efficiency.Solutions that satisfy application objectives while extending network life are needed to address severe ...In Wireless Sensor Networks(WSNs),survivability is a crucial issue that is greatly impacted by energy efficiency.Solutions that satisfy application objectives while extending network life are needed to address severe energy constraints inWSNs.This paper presents an Adaptive Enhanced GreyWolf Optimizer(AEGWO)for energy-efficient cluster head(CH)selection that mitigates the exploration–exploitation imbalance,preserves population diversity,and avoids premature convergence inherent in baseline GWO.The AEGWO combines adaptive control of the parameter of the search pressure to accelerate convergence without stagnation,a hybrid velocity-momentum update based on the dynamics of PSO,and an intelligent mutation operator to maintain the diversity of the population.The search is guided by a multi-objective fitness,which aims at maximizing the residual energy,equal distribution of CH,minimizing the intra-cluster distance,desirable proximity to sinks,and enhancing the coverage.Simulations on 100 nodes homogeneousWSN Tested the proposed AEGWO under the same conditions with LEACH,GWO,IGWO,PSO,WOA,and GA,AEGWO significantly increases stability and lifetime compared to LEACHand other tested algorithms;it has the best first,half,and last node dead,and higher residual energy and smaller communication overhead.The findings prove that AEGWO provides sustainable energy management and better lifetime extension,which makes it a robust,flexible clustering protocol of large-scaleWSNs.展开更多
Automated essay scoring(AES)systems have gained significant importance in educational settings,offering a scalable,efficient,and objective method for evaluating student essays.However,developing AES systems for Arabic...Automated essay scoring(AES)systems have gained significant importance in educational settings,offering a scalable,efficient,and objective method for evaluating student essays.However,developing AES systems for Arabic poses distinct challenges due to the language’s complex morphology,diglossia,and the scarcity of annotated datasets.This paper presents a hybrid approach to Arabic AES by combining text-based,vector-based,and embeddingbased similarity measures to improve essay scoring accuracy while minimizing the training data required.Using a large Arabic essay dataset categorized into thematic groups,the study conducted four experiments to evaluate the impact of feature selection,data size,and model performance.Experiment 1 established a baseline using a non-machine learning approach,selecting top-N correlated features to predict essay scores.The subsequent experiments employed 5-fold cross-validation.Experiment 2 showed that combining embedding-based,text-based,and vector-based features in a Random Forest(RF)model achieved an R2 of 88.92%and an accuracy of 83.3%within a 0.5-point tolerance.Experiment 3 further refined the feature selection process,demonstrating that 19 correlated features yielded optimal results,improving R2 to 88.95%.In Experiment 4,an optimal data efficiency training approach was introduced,where training data portions increased from 5%to 50%.The study found that using just 10%of the data achieved near-peak performance,with an R2 of 85.49%,emphasizing an effective trade-off between performance and computational costs.These findings highlight the potential of the hybrid approach for developing scalable Arabic AES systems,especially in low-resource environments,addressing linguistic challenges while ensuring efficient data usage.展开更多
We incorporate a non-Markovian feedback mechanism into the simulated bifurcation method for dynamical solvers addressing combinatorial optimization problems.By reinjecting a portion of dissipated kinetic energy into e...We incorporate a non-Markovian feedback mechanism into the simulated bifurcation method for dynamical solvers addressing combinatorial optimization problems.By reinjecting a portion of dissipated kinetic energy into each spin in a history-dependent and trajectory-informed manner,the method effectively suppresses early freezing induced by inelastic boundaries and enhances the system's ability to explore complex energy landscapes.Numerical results on the maximum cut(MAX-CUT)instances of fully connected Sherrington–Kirkpatrick(SK)spin glass models,including the 2000-spin K_(2000)benchmark,demonstrate that the non-Markovian algorithm significantly improves both solution quality and convergence speed.Tests on randomly generated SK instances with 100 to 1000 spins further indicate favorable scalability and substantial gains in computational efficiency.Moreover,the proposed scheme is well suited for massively parallel hardware implementations,such as field-programmable gate arrays,providing a practical and scalable approach for solving large-scale combinatorial optimization problems.展开更多
Given the growing number of vehicle accidents caused by unintended acceleration and braking failure,verifying Sudden Unintended Acceleration(SUA)incidents has become a persistent challenge.A central issue of debate is...Given the growing number of vehicle accidents caused by unintended acceleration and braking failure,verifying Sudden Unintended Acceleration(SUA)incidents has become a persistent challenge.A central issue of debate is whether such events stem frommechanical malfunctions or driver pedalmisapplications.However,existing verification procedures implemented by vehiclemanufacturers often involve closed tests after vehicle recalls;thus raising ongoing concerns about reliability and transparency.Consequently,there is a growing need for a user-driven framework that enables independent data acquisition and verification.Although previous studies have addressed SUA detection using deep learning,few have explored howclass granularity optimization affects power efficiency and inference performance in real-time Edge AI systems.To address this problem,this work presents a cloud-assisted artificial intelligence(AI)solution for the reliable verification of SUA occurrences.The proposed system integrates multimodal sensor streams including camera-based foot images,On-Board Diagnostics II(OBD-II)signals,and six-axismeasurements to determine whether the brake pedal was actually engaged at themoment of a suspected SUA.Beyond image acquisition,convolutional neural network(CNN)models perform real-time inference to classify the driver’s pedal operation states with the resulting outputs transmitted and archived in the cloud.A dedicated dataset of brake and accelerator pedal images was collected from 15 vehicles produced by 6 domestic and international manufacturers.Using this dataset,transfer learning techniques were applied to compare and analyze model performance and generalization as the CNN class granularity varied from coarse to fine levels.Furthermore,classification performance was evaluated in terms of latency and power efficiency under different class configurations.The experimental results demonstrated that the proposed solution identified the driver’s pedal behavior accurately and promptly,with the two-class model achieving the highest F1-score and accuracy among all granularity settings.展开更多
We propose an optimization method based on evolutionary computation for the design of broadband high-efficiency current-biased reverse load-modulation power amplifiers(CB-RLM PAs).First,given the reverse load-modulati...We propose an optimization method based on evolutionary computation for the design of broadband high-efficiency current-biased reverse load-modulation power amplifiers(CB-RLM PAs).First,given the reverse load-modulation characteristics of CB-RLM PAs,a comprehensive objective function is proposed that combines multi-state impedance trajectory constraints with in-band performance deviations.For the saturation and 6 dB power back-off(PBO)states,approximately optimal impedance regions on the Smith chart are derived using impedance constraint circles based on load-pull simulations.These regions are used together with in-band performance deviations(e.g.,saturated efficiency,6 dB PBO efficiency,and saturated output power)for matching network optimization and design.Second,a multi-objective evolutionary algorithm based on decomposition with adaptive weights,neighborhood,and global replacement is integrated with harmonic balance simulations to optimize design parameters and evaluate performance.Finally,to validate the proposed method,a broadband CB-RLM PA operating from 0.6 to 1.8 GHz is designed and fabricated.Measurement results show that the efficiencies at saturation,6 dB PBO,and 8 dB PBO all exceed 43.6%,with saturated output power being maintained at 40.9–41.5 dBm,which confirms the feasibility and effectiveness of the proposed broadband high-efficiency CB-RLM PA optimization and design approach.展开更多
The goal of the present work is to demonstrate the potential of Artificial Neural Network(ANN)-driven Genetic Algorithm(GA)methods for energy efficiency and economic performance optimization of energy efficiency measu...The goal of the present work is to demonstrate the potential of Artificial Neural Network(ANN)-driven Genetic Algorithm(GA)methods for energy efficiency and economic performance optimization of energy efficiency measures in a multi-family house building in Greece.The energy efficiency measures include different heating/cooling systems(such as low-temperature and high-temperature heat pumps,natural gas boilers,split units),building envelope components for floor,walls,roof and windows of variable heat transfer coefficients,the installation of solar thermal collectors and PVs.The calculations of the building loads and investment and operating and maintenance costs of the measures are based on the methodology defined in Directive 2010/31/EU,while economic assumptions are based on EN 15459-1 standard.Typically,multi-objective optimization of energy efficiency measures often requires the simulation of very large numbers of cases involving numerous possible combinations,resulting in intense computational load.The results of the study indicate that ANN-driven GA methods can be used as an alternative,valuable tool for reliably predicting the optimal measures which minimize primary energy consumption and life cycle cost of the building with greatly reduced computational requirements.Through GA methods,the computational time needed for obtaining the optimal solutions is reduced by 96.4%-96.8%.展开更多
The real-time AC optimal power flow(OPF)problem is a key issue in making fast and accurate decisions to ensure the safety and economy of power systems.With the rapid development of renewable energies,the fluctuation h...The real-time AC optimal power flow(OPF)problem is a key issue in making fast and accurate decisions to ensure the safety and economy of power systems.With the rapid development of renewable energies,the fluctuation has grown more vibrant,thus a novel approach called safe deep reinforcement learning is proposed in this paper.Herein,the real-time ACOPF problem is modeled as a constrained Markov decision process,and primal-dual optimization(PDO)based proximal policy optimization(PPO)is used to learn the optimal generator outputs in the primal domain and security constraints in the dual domain,which avoids manually selecting a trade-off between penalties for constraint violations and rewards for the economy.Before training,behavior cloning clones the expert experience into the initial weights of neural networks.Moreover,multiprocessing training is utilized to accelerate the training speed.Case studies are conducted on the IEEE 118-bus system and the modified IEEE 118-bus system.Compared with other methods,the experimental results show that the proposed method can achieve security and near-optimal economic goals by fast calculating the real-time ACOPF problem.展开更多
In this paper,we consider the maximal positive definite solution of the nonlinear matrix equation.By using the idea of Algorithm 2.1 in ZHANG(2013),a new inversion-free method with a stepsize parameter is proposed to ...In this paper,we consider the maximal positive definite solution of the nonlinear matrix equation.By using the idea of Algorithm 2.1 in ZHANG(2013),a new inversion-free method with a stepsize parameter is proposed to obtain the maximal positive definite solution of nonlinear matrix equation X+A^(*)X|^(-α)A=Q with the case 0<α≤1.Based on this method,a new iterative algorithm is developed,and its convergence proof is given.Finally,two numerical examples are provided to show the effectiveness of the proposed method.展开更多
With the large-scale integration of new energy sources,various resources such as energy storage,electric vehicles(EVs),and photovoltaics(PV) have participated in the scheduling of active distribution networks(ADNs),po...With the large-scale integration of new energy sources,various resources such as energy storage,electric vehicles(EVs),and photovoltaics(PV) have participated in the scheduling of active distribution networks(ADNs),posing new challenges to the operation and scheduling of distribution networks.Aiming at the uncertainty of PV and EV,an optimal scheduling model for ADNs based on multi-scenario fuzzy set based charging station resource forecasting is constructed.To address the scheduling uncertainties caused by PV and load forecasting errors,a day-ahead optimal scheduling model based on conditional value at risk(CVaR) for cost assessment is established,with the optimization objectives of minimizing the operation cost of distribution networks and the risk cost caused by forecasting errors.An improved subtractive optimizer algorithm is proposed to solve the model and formulate day-ahead optimization schemes.Secondly,a forecasting model for dispatchable resources in charging stations is constructed based on event-based fuzzy set theory.On this basis,an intraday scheduling model is built to comprehensively utilize the dispatchable resources of charging stations to coordinate with the output of distributed power sources,achieving optimal scheduling with the goal of minimizing operation costs.Finally,an experimental scenario based on the IEEE-33 node system is designed for simulation verification.The comparison of optimal scheduling results shows that the proposed method can fully exploit the potential scheduling resources of charging stations,improving the operation stability of ADNs and the accommodution capacity of new energy.展开更多
Organic room-temperature phosphorescence(RTP)materials are promising for bioimaging applications due to their tunable structures,excellent biocompatibility,and long-lived luminescence.However,the development of highly...Organic room-temperature phosphorescence(RTP)materials are promising for bioimaging applications due to their tunable structures,excellent biocompatibility,and long-lived luminescence.However,the development of highly efficient organic RTP materials for aqueous systems remains challenging,as the organic phosphorescence is prone to being quenched by the dissolved oxygen in water.Herein,heteroaromatic carboxylic acids serve as ligand vips to construct a series of host-vip composites with nontoxic,dense EDTA-M(M=Ca,Mg,and Al)coordination polymer in water.These composites exhibit ultra-long pure RTP of vip molecules with phosphorescence quantum yield up to 53%,and lifetime up to 589.7 ms,due to the synergistic effect of dual-network structure:a coordinatively cross-linked network of EDTA-M,and a non-covalent bonded network formed by ligands and water molecules.The phosphorescence intensity is more than three times that of the composite with a single coordination network.Notably,the dual-network configuration can form a rigid and dense structure and block the intrusion of external H_(2)O and O_(2) molecules to avoid phosphorescence quenching in water.As a result,the RTP of the composites remains unchanged after 1 month in water.Furthermore,the nanoparticles fabricated from composites and anionic surfactants can be successfully applied in in vivo imaging of mice for the stable RTP in water.This work provides a novel strategy for the development of high-performance RTP materials in aqueous systems.展开更多
Although the certified power conversion efficiency(PCE)of single-junction perovskite solar cells(PSCs)has achieved a high level of 27%,approaching the single-crystalline silicon solar cells,the device stability remain...Although the certified power conversion efficiency(PCE)of single-junction perovskite solar cells(PSCs)has achieved a high level of 27%,approaching the single-crystalline silicon solar cells,the device stability remains an urgent issue to be resolved for the commercialization.Defect passivation emerged as a viable approach to enhance the operational stability of the solar devices.Herein,phenylthiourea(PhTu)derivatives are selected as effective passivation agents to enhance the optoelectronic properties of printed methylammonium lead iodide(MAPbI_(3))films.It is demonstrated that incorporating a small amount of 1-(4-carboxyphenyl)-2-thiourea(PhTu-COOH)significantly reduces the trap-state density and leads to longer carrier lifetime of the perovskite films.As a result,the inverted solar device made of Ph Tu-COOH-modified MAPbI_(3) perovskite film shows remarkably improved efficiency(from 17.29%to 20.22%)and obviously increased open-circuit voltage(V_(OC))(from 1.043 to 1.143 V),as compared with the pristine device.Moreover,the Ph Tu-COOH-modified PSCs exhibit enhanced operational stability due to the significantly reduced trap-state density.Finally,the optimized solar module fabricated with an active area of 11.28 cm^(2) delivers a high PCE of 17.07%with negligible V_(OC)loss,demonstrating the feasibility of the blade-coating method for large-area perovskite film deposition.展开更多
Coal-derived hard carbon(HC)represents a promising anode material for sodium-ion batteries owing to its cost-effectiveness and high carbon yield.However,conventional carbonization induces excessive graphitization,yiel...Coal-derived hard carbon(HC)represents a promising anode material for sodium-ion batteries owing to its cost-effectiveness and high carbon yield.However,conventional carbonization induces excessive graphitization,yielding insufficient interlayer spacing(d_(002)<0.37 nm)and underdeveloped closed pores.Herein,we propose a dynamic crystallization control strategy through carbothermal shock treatment(1300°C,30 s)that decouples thermodynamic and kinetic constraints.This method precisely modulates graphite domain ordering kinetics,producing short-range ordered structures with expanded interlayer spacing(d_(002)=0.385 nm)and homogeneously distributed closed nanopores.Through combined in situ characterization and first-principles calculations,we elucidate a three-stage crystallization mechanism:(i)amorphous carbon transformation,(ii)open-pore collapse,and(iii)pseudo-graphitic ordering.The optimized HC achieves record performance with 88.6%initial Coulombic efficiency and 204 mA h g^(−1)plateau capacity,while its optimal interlayer spacing lowers Na+diffusion barriers to enable exceptional rate capability(221 mA h g^(−1)at 0.5C after 300 cycles).Practical pouch cells maintain 85%capacity retention after 100 cycles at−20°C and deliver 284 Wh kg^(−1)energy density.This work establishes a kinetic regulation paradigm for graphitization-prone precursors,advancing the rational design of high-performance HC anodes.展开更多
Efficient energy utilization in covert communication sustains covertness while assuring communication quality and efficiency.This paper investigates covert communication energy efficiency(EE)in direct uplink satellite...Efficient energy utilization in covert communication sustains covertness while assuring communication quality and efficiency.This paper investigates covert communication energy efficiency(EE)in direct uplink satellite-ground communications,focusing on enhancing system EE via optimized transmit beamforming and satellite orbit altitude selection.This paper first establishes an optimization problem to maximize system EE in a direct uplink satelliteground covert communication scenario.To solve this non-convex optimization problem,it is decomposed into two subproblems and solved using the successive convex approximation(SCA)method.Based on the above methods,this paper proposes an overall iterative optimization algorithm.Simulation results demonstrate that the proposed algorithm surpasses the conventional baseline algorithms in terms of system EE.Furthermore,they elucidate the correlation between the amount of information received by the receiver and the variations in the satellite’s orbital altitude.展开更多
With the increasing complexity of the current electromagnetic environment,excessive microwave radi-ation not only does harm to human health but also forms various electromagnetic interference to so-phisticated electro...With the increasing complexity of the current electromagnetic environment,excessive microwave radi-ation not only does harm to human health but also forms various electromagnetic interference to so-phisticated electronic instruments.Therefore,the design and preparation of electromagnetic absorbing composites represent an efficient approach to mitigate the current hazards of electromagnetic radiation.However,traditional electromagnetic absorbers are difficult to satisfy the demands of actual utilization in the face of new challenges,and emerging absorbents have garnered increasing attention due to their structure and performance-based advantages.In this review,several emerging composites of Mxene-based,biochar-based,chiral,and heat-resisting are discussed in detail,including their synthetic strategy,structural superiority and regulation method,and final optimization of electromagnetic absorption ca-pacity.These insights provide a comprehensive reference for the future development of new-generation electromagnetic-wave absorption composites.Moreover,the potential development directions of these emerging absorbers have been proposed as well.展开更多
An optimization model has been established and solved to determine the optimal threshold value for the event-triggered self-adaptive optimization strategy,which aims to strike a balance between optimization performanc...An optimization model has been established and solved to determine the optimal threshold value for the event-triggered self-adaptive optimization strategy,which aims to strike a balance between optimization performance and control load while ensuring continuous optimization.First,evaluation indicators are introduced to comprehensively analyze the impact of power fluctuations on the objective function and system voltage at both the system-wide and local levels.Based on these indicators,a multi-stage centralized optimization(MCO)is selectively applied,addressing system state deviations to achieve optimal operating states while maintaining a voltage security margin to ensure system safety.Then,distributed optimization(DO)is carried out at each bus with a renewable energy source or random load integration to accommodate short-term uncertainties using a self-adaptive reactive power algorithm.The optimal threshold value for event-triggered DO is calculated to balance control burden and optimization effectiveness.Utilizing the local state deviation evaluation indicator,unnecessary DOs are skipped when minor power fluctuations occur at the local level.Finally,following the linear superposition principle,event-triggered DOs executed at all distributed controllers collectively constitute the self-adaptive optimization strategy for the entire system.A case study on the IEEE New England 39-bus power system illustrates the effectiveness of the proposed strategy.展开更多
With the rapid growth of cloud computing,the number of data centers(DCs)continuously increases,leading to a high-energy consumption dilemma.Cooling,apart from IT equipment,represents the largest energy consumption in ...With the rapid growth of cloud computing,the number of data centers(DCs)continuously increases,leading to a high-energy consumption dilemma.Cooling,apart from IT equipment,represents the largest energy consumption in DCs.Passive design(PD)and active design(AD)are two important approaches in architectural design to reduce energy consumption.However,for DC cooling,few studies have summarized AD,and there are almost no studies on PD.Based on existing international research(2005-2024),this paper summarizes the current state of cooling strategies for DCs.PD encompasses floors,ceilings,and layout and zoning of racks.Additionally,other passive strategies not yet studied in DCs are critically examined.AD includes air,liquid,free,and two-phase cooling.This paper systematically compares the performance of different AD technologies on various KPIs,including energy,economic,and environmental indicators.This paper also explores the application of different cooling design strategies through best-practice examples and presents advanced algorithms for energy management in operational DCs.This study reveals that free cooling is widely employed,with Artificial Neural Networks emerging as the most popular algorithm for managing cooling energy.Finally,this paper suggests four future directions for reducing cooling energy in DCs,with a focus on the development of passive strategies.This paper provides an overview and guide to DC energy-consumption issues,emphasizes the importance of implementing passive and active design strategies to reduce DC cooling energy consumption,and provides directions and references for future energy-efficient DC designs.展开更多
In maize production,the development of density-tolerant and lodging-resistant varieties has made dense planting an effective strategy for achieving high and stable yields,with superior hybrids serving as a prerequisit...In maize production,the development of density-tolerant and lodging-resistant varieties has made dense planting an effective strategy for achieving high and stable yields,with superior hybrids serving as a prerequisite for successful highdensity cultivation.However,the photosynthetic mechanisms underlying improved density tolerance in maize hybrids released across different eras in China remain unclear.This study investigates 40 years of breeding progress toward enhanced photosynthetic traits under varying planting densities and elucidates the physiological and ecological bases of improved density tolerance in maize hybrids.A three-year field experiment was conducted from 2019 to 2021 to compare eight major Chinese hybrids from four decadal cohorts under three planting densities:45,000(D1),67,500(D2),and 90,000(D3)plants ha^(-1).At high density(D3),modern hybrids exhibited a more optimal canopy architecture and superior leaf photosynthetic performance compared to older hybrids,despite a slight reduction in specific leaf nitrogen.Notably,modern hybrids(2000s)were able to maintain higher net photosynthetic rates and photosynthetic nitrogen use efficiency(PNUE)at D3,resulting in the highest grain yield(GY),which was 118.47%greater than that of older hybrids(1970s).Leaf area duration after anthesis,total chlorophyll content,key photosynthetic enzyme activities,and maximum quantum efficiency of PSII photochemistry were all positively correlated with GY.Among these,PNUE showed the strongest correlation with grain yield and thus represents a key indicator for optimizing maize hybrids.Based on these findings,breeders should continue selecting hybrids under high-density and suboptimal conditions,focusing on optimizing population architecture and enhancing photosynthetic capacity while fine-tuning leaf nitrogen status to develop high-yielding,density-tolerant hybrids capable of sustaining long-term increases in maize grain yield.展开更多
Carbazole derivatives with a single phosphonic acid(PA)group are widely used as monolayer interfaces in perovskites and organic solar cells(OSCs).However,their hydrophilic nature renders ITO electrodes hydrophobic,lim...Carbazole derivatives with a single phosphonic acid(PA)group are widely used as monolayer interfaces in perovskites and organic solar cells(OSCs).However,their hydrophilic nature renders ITO electrodes hydrophobic,limiting further applications.In this study,a novel carbazole-based compound functionalized with two PA groups,denoted 2PACz-D1,was designed to create a dual hydrophilic interface.This configuration enables the formation of a bilayer hole-transporting layer(HTL).Specifically,one PA group anchors to the ITO electrode,while the other generates a secondary hydrophilic surface.This allows the subsequent deposition of hydrophilic PEDOT:PSS,forming a protective bilayer HTL that shields ITO from corrosive acidic polymers.The OSCs incorporating this bilayer HTL achieved a power conversion efficiency of 19.44%and exhibited improved thermal stability compared to devices with a single HTL.This work demonstrates the potential of bis-PA carbazole derivatives for tailoring the HTL surface properties,offering promising opportunities for various organic electronic devices.展开更多
In the era of massive data,the study of distributed data is a significant topic.Model averaging can be effectively applied to distributed data by combining information from all machines.For linear models,the model ave...In the era of massive data,the study of distributed data is a significant topic.Model averaging can be effectively applied to distributed data by combining information from all machines.For linear models,the model averaging approach has been developed in the context of distributed data.However,further investigation is needed for more complex models.In this paper,the authors propose a distributed optimal model averaging approach based on multivariate additive models,which approximates unknown functions using B-splines allowing each machine to have a different smoothing degree.To utilize the information from the covariance matrix of dependent errors in multivariate multiple regressions,the authors use the Mahalanobis distance to construct a Mallows-type weight choice criterion.The criterion can be computed by transmitting information between the local machines and the center machine in two steps.The authors demonstrate the asymptotic optimality of the proposed model averaging estimator when the covariates are subject to uncertainty,and obtain the convergence rate of the weight vector to the theoretically optimal weights.The results remain novel even for additive models with a single response variable.The numerical examples show that the proposed method yields good performance.展开更多
The personalized fine-tuning of large languagemodels(LLMs)on edge devices is severely constrained by limited computation resources.Although split federated learning alleviates on-device burdens,its effectiveness dimin...The personalized fine-tuning of large languagemodels(LLMs)on edge devices is severely constrained by limited computation resources.Although split federated learning alleviates on-device burdens,its effectiveness diminishes in few-shot reasoning scenarios due to the low data efficiency of conventional supervised fine-tuning,which leads to excessive communication overhead.To address this,we propose Language-Empowered Split Fine-Tuning(LESFT),a framework that integrates split architectures with a contrastive-inspired fine-tuning paradigm.LESFT simultaneously learns frommultiple logically equivalent but linguistically diverse reasoning chains,providing richer supervisory signals and improving data efficiency.This process-oriented training allows more effective reasoning adaptation with fewer samples.Extensive experiments demonstrate that LESFT consistently outperforms strong baselines such as SplitLoRA in task accuracy.LESFT consistently outperforms strong baselines on GSM8K,CommonsenseQA,and AQUA_RAT,with the largest gains observed on Qwen2.5-3B.These results indicate that LESFT can effectively adapt large language models for reasoning tasks under the computational and communication constraints of edge environments.展开更多
基金The Open Access publication fee for this article was fully covered by Abu Dhabi University.
文摘In Wireless Sensor Networks(WSNs),survivability is a crucial issue that is greatly impacted by energy efficiency.Solutions that satisfy application objectives while extending network life are needed to address severe energy constraints inWSNs.This paper presents an Adaptive Enhanced GreyWolf Optimizer(AEGWO)for energy-efficient cluster head(CH)selection that mitigates the exploration–exploitation imbalance,preserves population diversity,and avoids premature convergence inherent in baseline GWO.The AEGWO combines adaptive control of the parameter of the search pressure to accelerate convergence without stagnation,a hybrid velocity-momentum update based on the dynamics of PSO,and an intelligent mutation operator to maintain the diversity of the population.The search is guided by a multi-objective fitness,which aims at maximizing the residual energy,equal distribution of CH,minimizing the intra-cluster distance,desirable proximity to sinks,and enhancing the coverage.Simulations on 100 nodes homogeneousWSN Tested the proposed AEGWO under the same conditions with LEACH,GWO,IGWO,PSO,WOA,and GA,AEGWO significantly increases stability and lifetime compared to LEACHand other tested algorithms;it has the best first,half,and last node dead,and higher residual energy and smaller communication overhead.The findings prove that AEGWO provides sustainable energy management and better lifetime extension,which makes it a robust,flexible clustering protocol of large-scaleWSNs.
基金funded by Deanship of Graduate studies and Scientific Research at Jouf University under grant No.(DGSSR-2024-02-01264).
文摘Automated essay scoring(AES)systems have gained significant importance in educational settings,offering a scalable,efficient,and objective method for evaluating student essays.However,developing AES systems for Arabic poses distinct challenges due to the language’s complex morphology,diglossia,and the scarcity of annotated datasets.This paper presents a hybrid approach to Arabic AES by combining text-based,vector-based,and embeddingbased similarity measures to improve essay scoring accuracy while minimizing the training data required.Using a large Arabic essay dataset categorized into thematic groups,the study conducted four experiments to evaluate the impact of feature selection,data size,and model performance.Experiment 1 established a baseline using a non-machine learning approach,selecting top-N correlated features to predict essay scores.The subsequent experiments employed 5-fold cross-validation.Experiment 2 showed that combining embedding-based,text-based,and vector-based features in a Random Forest(RF)model achieved an R2 of 88.92%and an accuracy of 83.3%within a 0.5-point tolerance.Experiment 3 further refined the feature selection process,demonstrating that 19 correlated features yielded optimal results,improving R2 to 88.95%.In Experiment 4,an optimal data efficiency training approach was introduced,where training data portions increased from 5%to 50%.The study found that using just 10%of the data achieved near-peak performance,with an R2 of 85.49%,emphasizing an effective trade-off between performance and computational costs.These findings highlight the potential of the hybrid approach for developing scalable Arabic AES systems,especially in low-resource environments,addressing linguistic challenges while ensuring efficient data usage.
基金supported by the National Key Research and Development Program of China(Grant No.2024YFA1408500)the National Natural Science Foundation of China(Grant Nos.12174028 and 12574115)the Open Fund of the State Key Laboratory of Spintronics Devices and Technologies(Grant No.SPL-2408)。
文摘We incorporate a non-Markovian feedback mechanism into the simulated bifurcation method for dynamical solvers addressing combinatorial optimization problems.By reinjecting a portion of dissipated kinetic energy into each spin in a history-dependent and trajectory-informed manner,the method effectively suppresses early freezing induced by inelastic boundaries and enhances the system's ability to explore complex energy landscapes.Numerical results on the maximum cut(MAX-CUT)instances of fully connected Sherrington–Kirkpatrick(SK)spin glass models,including the 2000-spin K_(2000)benchmark,demonstrate that the non-Markovian algorithm significantly improves both solution quality and convergence speed.Tests on randomly generated SK instances with 100 to 1000 spins further indicate favorable scalability and substantial gains in computational efficiency.Moreover,the proposed scheme is well suited for massively parallel hardware implementations,such as field-programmable gate arrays,providing a practical and scalable approach for solving large-scale combinatorial optimization problems.
基金supported by Basic Science Research Program to Research Institute for Basic Sciences(RIBS)of Jeju National University through the National Research Foundation of Korea(NRF)funded by the Ministry of Education(RS-2019-NR040080)This research was also carried out with the support of the Jeju RISE Center,funded by the Ministry of Education and Jeju Special Self-Governing Province in 2025,as part of the“Regional Innovation System&Education(RISE):Glocal University 30”initiative.
文摘Given the growing number of vehicle accidents caused by unintended acceleration and braking failure,verifying Sudden Unintended Acceleration(SUA)incidents has become a persistent challenge.A central issue of debate is whether such events stem frommechanical malfunctions or driver pedalmisapplications.However,existing verification procedures implemented by vehiclemanufacturers often involve closed tests after vehicle recalls;thus raising ongoing concerns about reliability and transparency.Consequently,there is a growing need for a user-driven framework that enables independent data acquisition and verification.Although previous studies have addressed SUA detection using deep learning,few have explored howclass granularity optimization affects power efficiency and inference performance in real-time Edge AI systems.To address this problem,this work presents a cloud-assisted artificial intelligence(AI)solution for the reliable verification of SUA occurrences.The proposed system integrates multimodal sensor streams including camera-based foot images,On-Board Diagnostics II(OBD-II)signals,and six-axismeasurements to determine whether the brake pedal was actually engaged at themoment of a suspected SUA.Beyond image acquisition,convolutional neural network(CNN)models perform real-time inference to classify the driver’s pedal operation states with the resulting outputs transmitted and archived in the cloud.A dedicated dataset of brake and accelerator pedal images was collected from 15 vehicles produced by 6 domestic and international manufacturers.Using this dataset,transfer learning techniques were applied to compare and analyze model performance and generalization as the CNN class granularity varied from coarse to fine levels.Furthermore,classification performance was evaluated in terms of latency and power efficiency under different class configurations.The experimental results demonstrated that the proposed solution identified the driver’s pedal behavior accurately and promptly,with the two-class model achieving the highest F1-score and accuracy among all granularity settings.
基金supported by the National Natural Science Foundation of China(Nos.62171204,62171129,62001192).
文摘We propose an optimization method based on evolutionary computation for the design of broadband high-efficiency current-biased reverse load-modulation power amplifiers(CB-RLM PAs).First,given the reverse load-modulation characteristics of CB-RLM PAs,a comprehensive objective function is proposed that combines multi-state impedance trajectory constraints with in-band performance deviations.For the saturation and 6 dB power back-off(PBO)states,approximately optimal impedance regions on the Smith chart are derived using impedance constraint circles based on load-pull simulations.These regions are used together with in-band performance deviations(e.g.,saturated efficiency,6 dB PBO efficiency,and saturated output power)for matching network optimization and design.Second,a multi-objective evolutionary algorithm based on decomposition with adaptive weights,neighborhood,and global replacement is integrated with harmonic balance simulations to optimize design parameters and evaluate performance.Finally,to validate the proposed method,a broadband CB-RLM PA operating from 0.6 to 1.8 GHz is designed and fabricated.Measurement results show that the efficiencies at saturation,6 dB PBO,and 8 dB PBO all exceed 43.6%,with saturated output power being maintained at 40.9–41.5 dBm,which confirms the feasibility and effectiveness of the proposed broadband high-efficiency CB-RLM PA optimization and design approach.
文摘The goal of the present work is to demonstrate the potential of Artificial Neural Network(ANN)-driven Genetic Algorithm(GA)methods for energy efficiency and economic performance optimization of energy efficiency measures in a multi-family house building in Greece.The energy efficiency measures include different heating/cooling systems(such as low-temperature and high-temperature heat pumps,natural gas boilers,split units),building envelope components for floor,walls,roof and windows of variable heat transfer coefficients,the installation of solar thermal collectors and PVs.The calculations of the building loads and investment and operating and maintenance costs of the measures are based on the methodology defined in Directive 2010/31/EU,while economic assumptions are based on EN 15459-1 standard.Typically,multi-objective optimization of energy efficiency measures often requires the simulation of very large numbers of cases involving numerous possible combinations,resulting in intense computational load.The results of the study indicate that ANN-driven GA methods can be used as an alternative,valuable tool for reliably predicting the optimal measures which minimize primary energy consumption and life cycle cost of the building with greatly reduced computational requirements.Through GA methods,the computational time needed for obtaining the optimal solutions is reduced by 96.4%-96.8%.
基金supported by the National Natural Science Foundation of China(52007173 and U22B2098).
文摘The real-time AC optimal power flow(OPF)problem is a key issue in making fast and accurate decisions to ensure the safety and economy of power systems.With the rapid development of renewable energies,the fluctuation has grown more vibrant,thus a novel approach called safe deep reinforcement learning is proposed in this paper.Herein,the real-time ACOPF problem is modeled as a constrained Markov decision process,and primal-dual optimization(PDO)based proximal policy optimization(PPO)is used to learn the optimal generator outputs in the primal domain and security constraints in the dual domain,which avoids manually selecting a trade-off between penalties for constraint violations and rewards for the economy.Before training,behavior cloning clones the expert experience into the initial weights of neural networks.Moreover,multiprocessing training is utilized to accelerate the training speed.Case studies are conducted on the IEEE 118-bus system and the modified IEEE 118-bus system.Compared with other methods,the experimental results show that the proposed method can achieve security and near-optimal economic goals by fast calculating the real-time ACOPF problem.
基金Supported in part by Natural Science Foundation of Guangxi(2023GXNSFAA026246)in part by the Central Government's Guide to Local Science and Technology Development Fund(GuikeZY23055044)in part by the National Natural Science Foundation of China(62363003)。
文摘In this paper,we consider the maximal positive definite solution of the nonlinear matrix equation.By using the idea of Algorithm 2.1 in ZHANG(2013),a new inversion-free method with a stepsize parameter is proposed to obtain the maximal positive definite solution of nonlinear matrix equation X+A^(*)X|^(-α)A=Q with the case 0<α≤1.Based on this method,a new iterative algorithm is developed,and its convergence proof is given.Finally,two numerical examples are provided to show the effectiveness of the proposed method.
基金Supported by the Technology Project of State Grid Corporation Headquarters(No.5100-202322029A-1-1-ZN)the 2024 Youth Science Foundation Project of China (No.62303006)。
文摘With the large-scale integration of new energy sources,various resources such as energy storage,electric vehicles(EVs),and photovoltaics(PV) have participated in the scheduling of active distribution networks(ADNs),posing new challenges to the operation and scheduling of distribution networks.Aiming at the uncertainty of PV and EV,an optimal scheduling model for ADNs based on multi-scenario fuzzy set based charging station resource forecasting is constructed.To address the scheduling uncertainties caused by PV and load forecasting errors,a day-ahead optimal scheduling model based on conditional value at risk(CVaR) for cost assessment is established,with the optimization objectives of minimizing the operation cost of distribution networks and the risk cost caused by forecasting errors.An improved subtractive optimizer algorithm is proposed to solve the model and formulate day-ahead optimization schemes.Secondly,a forecasting model for dispatchable resources in charging stations is constructed based on event-based fuzzy set theory.On this basis,an intraday scheduling model is built to comprehensively utilize the dispatchable resources of charging stations to coordinate with the output of distributed power sources,achieving optimal scheduling with the goal of minimizing operation costs.Finally,an experimental scenario based on the IEEE-33 node system is designed for simulation verification.The comparison of optimal scheduling results shows that the proposed method can fully exploit the potential scheduling resources of charging stations,improving the operation stability of ADNs and the accommodution capacity of new energy.
基金supported by the Startup Funds for Introduced Talents of Wuyi University(YJ202304)the National Natural Science Foundation of China(22375044).
文摘Organic room-temperature phosphorescence(RTP)materials are promising for bioimaging applications due to their tunable structures,excellent biocompatibility,and long-lived luminescence.However,the development of highly efficient organic RTP materials for aqueous systems remains challenging,as the organic phosphorescence is prone to being quenched by the dissolved oxygen in water.Herein,heteroaromatic carboxylic acids serve as ligand vips to construct a series of host-vip composites with nontoxic,dense EDTA-M(M=Ca,Mg,and Al)coordination polymer in water.These composites exhibit ultra-long pure RTP of vip molecules with phosphorescence quantum yield up to 53%,and lifetime up to 589.7 ms,due to the synergistic effect of dual-network structure:a coordinatively cross-linked network of EDTA-M,and a non-covalent bonded network formed by ligands and water molecules.The phosphorescence intensity is more than three times that of the composite with a single coordination network.Notably,the dual-network configuration can form a rigid and dense structure and block the intrusion of external H_(2)O and O_(2) molecules to avoid phosphorescence quenching in water.As a result,the RTP of the composites remains unchanged after 1 month in water.Furthermore,the nanoparticles fabricated from composites and anionic surfactants can be successfully applied in in vivo imaging of mice for the stable RTP in water.This work provides a novel strategy for the development of high-performance RTP materials in aqueous systems.
基金supported by the National Natural Science Foundation of China(Grant No.62205103)the Natural Science Foundation of Hunan Province(Grant No.2023JJ40216)the Elite Youth Program by the Department of Education of Hunan Province(Grant No.24B0663)。
文摘Although the certified power conversion efficiency(PCE)of single-junction perovskite solar cells(PSCs)has achieved a high level of 27%,approaching the single-crystalline silicon solar cells,the device stability remains an urgent issue to be resolved for the commercialization.Defect passivation emerged as a viable approach to enhance the operational stability of the solar devices.Herein,phenylthiourea(PhTu)derivatives are selected as effective passivation agents to enhance the optoelectronic properties of printed methylammonium lead iodide(MAPbI_(3))films.It is demonstrated that incorporating a small amount of 1-(4-carboxyphenyl)-2-thiourea(PhTu-COOH)significantly reduces the trap-state density and leads to longer carrier lifetime of the perovskite films.As a result,the inverted solar device made of Ph Tu-COOH-modified MAPbI_(3) perovskite film shows remarkably improved efficiency(from 17.29%to 20.22%)and obviously increased open-circuit voltage(V_(OC))(from 1.043 to 1.143 V),as compared with the pristine device.Moreover,the Ph Tu-COOH-modified PSCs exhibit enhanced operational stability due to the significantly reduced trap-state density.Finally,the optimized solar module fabricated with an active area of 11.28 cm^(2) delivers a high PCE of 17.07%with negligible V_(OC)loss,demonstrating the feasibility of the blade-coating method for large-area perovskite film deposition.
基金supported by the Key Laboratory of Sichuan Province for Lithium Resources Comprehensive Utilization and New Lithium Based Materials for Advanced Battery Technology(LRMKF202405)the National Natural Science Foundation of China(52402226)the Natural Science Foundation of Sichuan Province(2024NSFSC1016).
文摘Coal-derived hard carbon(HC)represents a promising anode material for sodium-ion batteries owing to its cost-effectiveness and high carbon yield.However,conventional carbonization induces excessive graphitization,yielding insufficient interlayer spacing(d_(002)<0.37 nm)and underdeveloped closed pores.Herein,we propose a dynamic crystallization control strategy through carbothermal shock treatment(1300°C,30 s)that decouples thermodynamic and kinetic constraints.This method precisely modulates graphite domain ordering kinetics,producing short-range ordered structures with expanded interlayer spacing(d_(002)=0.385 nm)and homogeneously distributed closed nanopores.Through combined in situ characterization and first-principles calculations,we elucidate a three-stage crystallization mechanism:(i)amorphous carbon transformation,(ii)open-pore collapse,and(iii)pseudo-graphitic ordering.The optimized HC achieves record performance with 88.6%initial Coulombic efficiency and 204 mA h g^(−1)plateau capacity,while its optimal interlayer spacing lowers Na+diffusion barriers to enable exceptional rate capability(221 mA h g^(−1)at 0.5C after 300 cycles).Practical pouch cells maintain 85%capacity retention after 100 cycles at−20°C and deliver 284 Wh kg^(−1)energy density.This work establishes a kinetic regulation paradigm for graphitization-prone precursors,advancing the rational design of high-performance HC anodes.
基金supported in part by the National Natural Science Foundation of China under Grants 62025110,62271093sponsored by Natural Science Foundation of Chongqing,China,under Grant CSTB2023NSCQ-LZX0108.
文摘Efficient energy utilization in covert communication sustains covertness while assuring communication quality and efficiency.This paper investigates covert communication energy efficiency(EE)in direct uplink satellite-ground communications,focusing on enhancing system EE via optimized transmit beamforming and satellite orbit altitude selection.This paper first establishes an optimization problem to maximize system EE in a direct uplink satelliteground covert communication scenario.To solve this non-convex optimization problem,it is decomposed into two subproblems and solved using the successive convex approximation(SCA)method.Based on the above methods,this paper proposes an overall iterative optimization algorithm.Simulation results demonstrate that the proposed algorithm surpasses the conventional baseline algorithms in terms of system EE.Furthermore,they elucidate the correlation between the amount of information received by the receiver and the variations in the satellite’s orbital altitude.
基金supported by the Surface Project of Local De-velopment in Science and Technology Guided by Central Govern-ment(No.2021ZYD0041)the National Natural Science Founda-tion of China(Nos.52377026 and 52301192)+3 种基金the Natural Science Foundation of Shandong Province(No.ZR2019YQ24)the Taishan Scholars and Young Experts Program of Shandong Province(No.tsqn202103057)the Special Financial of Shandong Province(Struc-tural Design of High-efficiency Electromagnetic Wave-absorbing Composite Materials and Construction of Shandong Provincial Tal-ent Teams)the“Sanqin Scholars”Innovation Teams Project of Shaanxi Province(Clean Energy Materials and High-Performance Devices Innovation Team of Shaanxi Dongling Smelting Co.,Ltd.).
文摘With the increasing complexity of the current electromagnetic environment,excessive microwave radi-ation not only does harm to human health but also forms various electromagnetic interference to so-phisticated electronic instruments.Therefore,the design and preparation of electromagnetic absorbing composites represent an efficient approach to mitigate the current hazards of electromagnetic radiation.However,traditional electromagnetic absorbers are difficult to satisfy the demands of actual utilization in the face of new challenges,and emerging absorbents have garnered increasing attention due to their structure and performance-based advantages.In this review,several emerging composites of Mxene-based,biochar-based,chiral,and heat-resisting are discussed in detail,including their synthetic strategy,structural superiority and regulation method,and final optimization of electromagnetic absorption ca-pacity.These insights provide a comprehensive reference for the future development of new-generation electromagnetic-wave absorption composites.Moreover,the potential development directions of these emerging absorbers have been proposed as well.
基金supported in part by National Key R&D Program of China(2022YFF0610600).
文摘An optimization model has been established and solved to determine the optimal threshold value for the event-triggered self-adaptive optimization strategy,which aims to strike a balance between optimization performance and control load while ensuring continuous optimization.First,evaluation indicators are introduced to comprehensively analyze the impact of power fluctuations on the objective function and system voltage at both the system-wide and local levels.Based on these indicators,a multi-stage centralized optimization(MCO)is selectively applied,addressing system state deviations to achieve optimal operating states while maintaining a voltage security margin to ensure system safety.Then,distributed optimization(DO)is carried out at each bus with a renewable energy source or random load integration to accommodate short-term uncertainties using a self-adaptive reactive power algorithm.The optimal threshold value for event-triggered DO is calculated to balance control burden and optimization effectiveness.Utilizing the local state deviation evaluation indicator,unnecessary DOs are skipped when minor power fluctuations occur at the local level.Finally,following the linear superposition principle,event-triggered DOs executed at all distributed controllers collectively constitute the self-adaptive optimization strategy for the entire system.A case study on the IEEE New England 39-bus power system illustrates the effectiveness of the proposed strategy.
文摘With the rapid growth of cloud computing,the number of data centers(DCs)continuously increases,leading to a high-energy consumption dilemma.Cooling,apart from IT equipment,represents the largest energy consumption in DCs.Passive design(PD)and active design(AD)are two important approaches in architectural design to reduce energy consumption.However,for DC cooling,few studies have summarized AD,and there are almost no studies on PD.Based on existing international research(2005-2024),this paper summarizes the current state of cooling strategies for DCs.PD encompasses floors,ceilings,and layout and zoning of racks.Additionally,other passive strategies not yet studied in DCs are critically examined.AD includes air,liquid,free,and two-phase cooling.This paper systematically compares the performance of different AD technologies on various KPIs,including energy,economic,and environmental indicators.This paper also explores the application of different cooling design strategies through best-practice examples and presents advanced algorithms for energy management in operational DCs.This study reveals that free cooling is widely employed,with Artificial Neural Networks emerging as the most popular algorithm for managing cooling energy.Finally,this paper suggests four future directions for reducing cooling energy in DCs,with a focus on the development of passive strategies.This paper provides an overview and guide to DC energy-consumption issues,emphasizes the importance of implementing passive and active design strategies to reduce DC cooling energy consumption,and provides directions and references for future energy-efficient DC designs.
基金supported by the National Natural Science Foundation of China(32071960)the National Key Research and Development Program of China(2018YFD0300603)。
文摘In maize production,the development of density-tolerant and lodging-resistant varieties has made dense planting an effective strategy for achieving high and stable yields,with superior hybrids serving as a prerequisite for successful highdensity cultivation.However,the photosynthetic mechanisms underlying improved density tolerance in maize hybrids released across different eras in China remain unclear.This study investigates 40 years of breeding progress toward enhanced photosynthetic traits under varying planting densities and elucidates the physiological and ecological bases of improved density tolerance in maize hybrids.A three-year field experiment was conducted from 2019 to 2021 to compare eight major Chinese hybrids from four decadal cohorts under three planting densities:45,000(D1),67,500(D2),and 90,000(D3)plants ha^(-1).At high density(D3),modern hybrids exhibited a more optimal canopy architecture and superior leaf photosynthetic performance compared to older hybrids,despite a slight reduction in specific leaf nitrogen.Notably,modern hybrids(2000s)were able to maintain higher net photosynthetic rates and photosynthetic nitrogen use efficiency(PNUE)at D3,resulting in the highest grain yield(GY),which was 118.47%greater than that of older hybrids(1970s).Leaf area duration after anthesis,total chlorophyll content,key photosynthetic enzyme activities,and maximum quantum efficiency of PSII photochemistry were all positively correlated with GY.Among these,PNUE showed the strongest correlation with grain yield and thus represents a key indicator for optimizing maize hybrids.Based on these findings,breeders should continue selecting hybrids under high-density and suboptimal conditions,focusing on optimizing population architecture and enhancing photosynthetic capacity while fine-tuning leaf nitrogen status to develop high-yielding,density-tolerant hybrids capable of sustaining long-term increases in maize grain yield.
基金supported by the National Key Research and Development Program of China(No.2022YFB4200400)the National Natural Science Foundation of China(Nos.W2511056,52503289 and 52333005)+1 种基金Beijing Natural Science Foundation(No.Z230018)the Academic Excellence Foundation of BUAA for PhD Students。
文摘Carbazole derivatives with a single phosphonic acid(PA)group are widely used as monolayer interfaces in perovskites and organic solar cells(OSCs).However,their hydrophilic nature renders ITO electrodes hydrophobic,limiting further applications.In this study,a novel carbazole-based compound functionalized with two PA groups,denoted 2PACz-D1,was designed to create a dual hydrophilic interface.This configuration enables the formation of a bilayer hole-transporting layer(HTL).Specifically,one PA group anchors to the ITO electrode,while the other generates a secondary hydrophilic surface.This allows the subsequent deposition of hydrophilic PEDOT:PSS,forming a protective bilayer HTL that shields ITO from corrosive acidic polymers.The OSCs incorporating this bilayer HTL achieved a power conversion efficiency of 19.44%and exhibited improved thermal stability compared to devices with a single HTL.This work demonstrates the potential of bis-PA carbazole derivatives for tailoring the HTL surface properties,offering promising opportunities for various organic electronic devices.
基金supported by Youth Academic Innocation Team Construction project of Capital University of Economics and Business under Grant No.QNTD202303supported by the Beijing Outstanding Young Scientist Program under Grant No.JWZQ20240101027the National Natural Science Foundation of China under Grant Nos.12031016,12531012 and 12426308。
文摘In the era of massive data,the study of distributed data is a significant topic.Model averaging can be effectively applied to distributed data by combining information from all machines.For linear models,the model averaging approach has been developed in the context of distributed data.However,further investigation is needed for more complex models.In this paper,the authors propose a distributed optimal model averaging approach based on multivariate additive models,which approximates unknown functions using B-splines allowing each machine to have a different smoothing degree.To utilize the information from the covariance matrix of dependent errors in multivariate multiple regressions,the authors use the Mahalanobis distance to construct a Mallows-type weight choice criterion.The criterion can be computed by transmitting information between the local machines and the center machine in two steps.The authors demonstrate the asymptotic optimality of the proposed model averaging estimator when the covariates are subject to uncertainty,and obtain the convergence rate of the weight vector to the theoretically optimal weights.The results remain novel even for additive models with a single response variable.The numerical examples show that the proposed method yields good performance.
基金supported in part by the National Natural Science Foundation of China(NSFC)under Grant 62276109The authors extend their appreciation to the Deanship of Scientific Research at King Saud University for funding this work through the Research Group Project number(ORF-2025-585).
文摘The personalized fine-tuning of large languagemodels(LLMs)on edge devices is severely constrained by limited computation resources.Although split federated learning alleviates on-device burdens,its effectiveness diminishes in few-shot reasoning scenarios due to the low data efficiency of conventional supervised fine-tuning,which leads to excessive communication overhead.To address this,we propose Language-Empowered Split Fine-Tuning(LESFT),a framework that integrates split architectures with a contrastive-inspired fine-tuning paradigm.LESFT simultaneously learns frommultiple logically equivalent but linguistically diverse reasoning chains,providing richer supervisory signals and improving data efficiency.This process-oriented training allows more effective reasoning adaptation with fewer samples.Extensive experiments demonstrate that LESFT consistently outperforms strong baselines such as SplitLoRA in task accuracy.LESFT consistently outperforms strong baselines on GSM8K,CommonsenseQA,and AQUA_RAT,with the largest gains observed on Qwen2.5-3B.These results indicate that LESFT can effectively adapt large language models for reasoning tasks under the computational and communication constraints of edge environments.
