This research introduces a unique approach to segmenting breast cancer images using a U-Net-based architecture.However,the computational demand for image processing is very high.Therefore,we have conducted this resear...This research introduces a unique approach to segmenting breast cancer images using a U-Net-based architecture.However,the computational demand for image processing is very high.Therefore,we have conducted this research to build a system that enables image segmentation training with low-power machines.To accomplish this,all data are divided into several segments,each being trained separately.In the case of prediction,the initial output is predicted from each trained model for an input,where the ultimate output is selected based on the pixel-wise majority voting of the expected outputs,which also ensures data privacy.In addition,this kind of distributed training system allows different computers to be used simultaneously.That is how the training process takes comparatively less time than typical training approaches.Even after completing the training,the proposed prediction system allows a newly trained model to be included in the system.Thus,the prediction is consistently more accurate.We evaluated the effectiveness of the ultimate output based on four performance matrices:average pixel accuracy,mean absolute error,average specificity,and average balanced accuracy.The experimental results show that the scores of average pixel accuracy,mean absolute error,average specificity,and average balanced accuracy are 0.9216,0.0687,0.9477,and 0.8674,respectively.In addition,the proposed method was compared with four other state-of-the-art models in terms of total training time and usage of computational resources.And it outperformed all of them in these aspects.展开更多
With the development of economic globalization,distributedmanufacturing is becomingmore andmore prevalent.Recently,integrated scheduling of distributed production and assembly has captured much concern.This research s...With the development of economic globalization,distributedmanufacturing is becomingmore andmore prevalent.Recently,integrated scheduling of distributed production and assembly has captured much concern.This research studies a distributed flexible job shop scheduling problem with assembly operations.Firstly,a mixed integer programming model is formulated to minimize the maximum completion time.Secondly,a Q-learning-assisted coevolutionary algorithmis presented to solve themodel:(1)Multiple populations are developed to seek required decisions simultaneously;(2)An encoding and decoding method based on problem features is applied to represent individuals;(3)A hybrid approach of heuristic rules and random methods is employed to acquire a high-quality population;(4)Three evolutionary strategies having crossover and mutation methods are adopted to enhance exploration capabilities;(5)Three neighborhood structures based on problem features are constructed,and a Q-learning-based iterative local search method is devised to improve exploitation abilities.The Q-learning approach is applied to intelligently select better neighborhood structures.Finally,a group of instances is constructed to perform comparison experiments.The effectiveness of the Q-learning approach is verified by comparing the developed algorithm with its variant without the Q-learning method.Three renowned meta-heuristic algorithms are used in comparison with the developed algorithm.The comparison results demonstrate that the designed method exhibits better performance in coping with the formulated problem.展开更多
The distributed flexible job shop scheduling problem(DFJSP)has attracted great attention with the growth of the global manufacturing industry.General DFJSP research only considers machine constraints and ignores worke...The distributed flexible job shop scheduling problem(DFJSP)has attracted great attention with the growth of the global manufacturing industry.General DFJSP research only considers machine constraints and ignores worker constraints.As one critical factor of production,effective utilization of worker resources can increase productivity.Meanwhile,energy consumption is a growing concern due to the increasingly serious environmental issues.Therefore,the distributed flexible job shop scheduling problem with dual resource constraints(DFJSP-DRC)for minimizing makespan and total energy consumption is studied in this paper.To solve the problem,we present a multi-objective mathematical model for DFJSP-DRC and propose a Q-learning-based multi-objective grey wolf optimizer(Q-MOGWO).In Q-MOGWO,high-quality initial solutions are generated by a hybrid initialization strategy,and an improved active decoding strategy is designed to obtain the scheduling schemes.To further enhance the local search capability and expand the solution space,two wolf predation strategies and three critical factory neighborhood structures based on Q-learning are proposed.These strategies and structures enable Q-MOGWO to explore the solution space more efficiently and thus find better Pareto solutions.The effectiveness of Q-MOGWO in addressing DFJSP-DRC is verified through comparison with four algorithms using 45 instances.The results reveal that Q-MOGWO outperforms comparison algorithms in terms of solution quality.展开更多
The flow shop scheduling problem is important for the manufacturing industry.Effective flow shop scheduling can bring great benefits to the industry.However,there are few types of research on Distributed Hybrid Flow S...The flow shop scheduling problem is important for the manufacturing industry.Effective flow shop scheduling can bring great benefits to the industry.However,there are few types of research on Distributed Hybrid Flow Shop Problems(DHFSP)by learning assisted meta-heuristics.This work addresses a DHFSP with minimizing the maximum completion time(Makespan).First,a mathematical model is developed for the concerned DHFSP.Second,four Q-learning-assisted meta-heuristics,e.g.,genetic algorithm(GA),artificial bee colony algorithm(ABC),particle swarm optimization(PSO),and differential evolution(DE),are proposed.According to the nature of DHFSP,six local search operations are designed for finding high-quality solutions in local space.Instead of randomselection,Q-learning assists meta-heuristics in choosing the appropriate local search operations during iterations.Finally,based on 60 cases,comprehensive numerical experiments are conducted to assess the effectiveness of the proposed algorithms.The experimental results and discussions prove that using Q-learning to select appropriate local search operations is more effective than the random strategy.To verify the competitiveness of the Q-learning assistedmeta-heuristics,they are compared with the improved iterated greedy algorithm(IIG),which is also for solving DHFSP.The Friedman test is executed on the results by five algorithms.It is concluded that the performance of four Q-learning-assisted meta-heuristics are better than IIG,and the Q-learning-assisted PSO shows the best competitiveness.展开更多
The emerging virtual coupling technology aims to operate multiple train units in a Virtually Coupled Train Set(VCTS)at a minimal but safe distance.To guarantee collision avoidance,the safety distance should be calcula...The emerging virtual coupling technology aims to operate multiple train units in a Virtually Coupled Train Set(VCTS)at a minimal but safe distance.To guarantee collision avoidance,the safety distance should be calculated using the state-of-the-art space-time separation principle that separates the Emergency Braking(EB)trajectories of two successive units during the whole EB process.In this case,the minimal safety distance is usually numerically calculated without an analytic formulation.Thus,the constrained VCTS control problem is hard to address with space-time separation,which is still a gap in the existing literature.To solve this problem,we propose a Distributed Economic Model Predictive Control(DEMPC)approach with computation efficiency and theoretical guarantee.Specifically,to alleviate the computation burden,we transform implicit safety constraints into explicitly linear ones,such that the optimal control problem in DEMPC is a quadratic programming problem that can be solved efficiently.For theoretical analysis,sufficient conditions are derived to guarantee the recursive feasibility and stability of DEMPC,employing compatibility constraints,tube techniques and terminal ingredient tuning.Moreover,we extend our approach with globally optimal and distributed online EB configuration methods to shorten the minimal distance among VCTS.Finally,experimental results demonstrate the performance and advantages of the proposed approaches.展开更多
In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.How...In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.However,the difference in the feature space between natural and adversarial examples hinders the accuracy and robustness of the model in adversarial training.This paper proposes a learnable distribution adversarial training method,aiming to construct the same distribution for training data utilizing the Gaussian mixture model.The distribution centroid is built to classify samples and constrain the distribution of the sample features.The natural and adversarial examples are pushed to the same distribution centroid to improve the accuracy and robustness of the model.The proposed method generates adversarial examples to close the distribution gap between the natural and adversarial examples through an attack algorithm explicitly designed for adversarial training.This algorithm gradually increases the accuracy and robustness of the model by scaling perturbation.Finally,the proposed method outputs the predicted labels and the distance between the sample and the distribution centroid.The distribution characteristics of the samples can be utilized to detect adversarial cases that can potentially evade the model defense.The effectiveness of the proposed method is demonstrated through comprehensive experiments.展开更多
The traditional distributed tactical training simulation system is limited by the availability and bandwidth of military network transmission channel and does not have the feasibility of remote interconnection and spa...The traditional distributed tactical training simulation system is limited by the availability and bandwidth of military network transmission channel and does not have the feasibility of remote interconnection and spatial sub-regional deployment.In this paper,a new communication method of distributed tactical training simulation system is proposed to solve the problem of strong business coupling between nodes and system availability under the condition of low bandwidth.The operator of federated exchange,federated queue and their concepts and design requirements are firstly proposed,and the inverted tree,triangle,ring exchange topology and circular queue structure are further constructed.Theoretically,the expected goal of high-speed interworking between nodes in the cluster and high reliable transmission between clusters is realized.The example also shows that this method can significantly improve the throughput of single switching node and federated node after using reliability confirmation mechanism.展开更多
How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classif...How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classification due to the powerful feature representation ability and better performance. However,the training and testing of CNN mainly rely on single machine.Single machine has its natural limitation and bottleneck in processing RSIs due to limited hardware resources and huge time consuming. Besides, overfitting is a challenge for the CNN model due to the unbalance between RSIs data and the model structure.When a model is complex or the training data is relatively small,overfitting occurs and leads to a poor predictive performance. To address these problems, a distributed CNN architecture for RSIs target classification is proposed, which dramatically increases the training speed of CNN and system scalability. It improves the storage ability and processing efficiency of RSIs. Furthermore,Bayesian regularization approach is utilized in order to initialize the weights of the CNN extractor, which increases the robustness and flexibility of the CNN model. It helps prevent the overfitting and avoid the local optima caused by limited RSI training images or the inappropriate CNN structure. In addition, considering the efficiency of the Na¨?ve Bayes classifier, a distributed Na¨?ve Bayes classifier is designed to reduce the training cost. Compared with other algorithms, the proposed system and method perform the best and increase the recognition accuracy. The results show that the distributed system framework and the proposed algorithms are suitable for RSIs target classification tasks.展开更多
The dynamic load distribution within in-service axlebox bearings of high-speed trains is crucial for the fatigue reliability assessment and forward design of axlebox bearings. This paper presents an in situ measuremen...The dynamic load distribution within in-service axlebox bearings of high-speed trains is crucial for the fatigue reliability assessment and forward design of axlebox bearings. This paper presents an in situ measurement of the dynamic load distribution in the four rows of two axlebox bearings on a bogie wheelset of a high-speed train under polygonal wheel–rail excitation. The measurement employed an improved strain-based method to measure the dynamic radial load distribution of roller bearings. The four rows of two axlebox bearings on a wheelset exhibited different ranges of loaded zones and different means of distributed loads. Besides, the mean value and standard deviation of measured roller–raceway contact loads showed non-monotonic variations with the frequency of wheel–rail excitation. The fatigue life of the four bearing rows under polygonal wheel–rail excitation was quantitatively predicted by compiling the measured roller–raceway contact load spectra of the most loaded position and considering the load spectra as input.展开更多
The current research on the aerodynamic performance of the train running in rainy weather is primarily concerned with the trajectory of the raindrops and the aerodynamic variation of trains caused by raindrops.In fact...The current research on the aerodynamic performance of the train running in rainy weather is primarily concerned with the trajectory of the raindrops and the aerodynamic variation of trains caused by raindrops.In fact,water film will generate on the train body when raindrops hit the train,which interacts with the flow field around the train,and would probably affect the aerodynamic performance of the train.In this paper,based on shear stress transport(SST)k-w turbulence model and Euler-Lagrange discrete phase model,the aerodynamic calculation model of a highspeed train under rainfall environment is established.The LWF(Lagrangian wall film)is used to simulate the water film distribution of the high-speed train under different rainfall intensities,and the aerodynamic performance of the train are studied.The calculation results show that raindrops will gather on the train surface and form water film under rainfall environment.With the extension of rainfall time,the thickness and coverage range of water film expand,and the maximum thickness of water film can reach 4.95 mm under the working conditions in this paper.The average thickness of water film on the train body increases with the rainfall intensity.When the rainfall intensity increases from 100 mm/h to 500 mm/h,the average water film thickness will increase by 3.26 times.The velocity of water film in the streamlined area of head car is larger than that in other areas,and the maximum velocity is 22.14 m/s.Compared with the rainless environment condition,the skin friction coefficient of the high-speed train increases and the average value will increase by 10.74%for a rainfall intensity of 500 mm/h.The positive pressure and resistance coefficient of the head car increase with the rainfall intensity.This research proposes a methodology to systematically analyze the generation of water film on the train surface and its influence on the train aerodynamic performance;the analysis can provide more practical results and can serve as a reference basis for the design and development of high-speed trains.展开更多
BACKGROUND： Elk-1 mRNA distributes extensively in the neurons of mice, rat and human brains, and the Elk-1 expression may be correlated with the synaptic plasticity, learning and memory. OBJECTIVE： To observe the di...BACKGROUND： Elk-1 mRNA distributes extensively in the neurons of mice, rat and human brains, and the Elk-1 expression may be correlated with the synaptic plasticity, learning and memory. OBJECTIVE： To observe the distribution of phosphorylated Elk-1 （pEIk-1） in whole brain of rats received Y-maze active avoidance training and the changes of pEIk-1 expression at different time points after training. DESIGN ： A randomized controlled study SETTING ： Research Room of Neurobiology, the Second Affiliated Hospital of Southern Medical University MATERIALS ： Fifty-five male clean-degree SD rats of 3-4 months old, weighing 200-250 g, were provided by the Experimental Animal Center of Southem Medical University. The rabbit anti-monoclonal pEIk-1 antibody was purchased from Cell Signal Transduction Company, and ABC kit from Vector Company. METHODS ： The experiments were carried out in the Research Room of Neurobiology, Second Affiliated Hospital of Southern Medical University from September 2004 to February 2005. ① Grouping： The rats were randomly divided into training group （n = 25）, sham-training group （n = 25） and normal control group （n = 5）, and the training and sham-training groups were observed at 0, 1, 3, 6 and 24 hours after training, which represented the five phases in the process of leaming and memory. ② Y-maze training： The rats were preconditioned in the electrical Y-maze apparatus, 20 minutes a day for 3 days continuously, and training began from the 4^th day. In the training group, the rats were trained with the combination of light and electddty. Each rat repeated for 60 times in each training, and the correct times were recorded, those correct for less than 25 times were taken as unqualified, and excluded from the training group, and supplemented by other rats in time. In the sham-training group, there was no fixed correlation between the application of light and electricity. The rats in the normal contrel group were given not any training. ③Detection of pEIk-1 expression： The rats were anesthetized after Y-maze training, brain tissue was removed to prepare coronal freezing sections, and the pEIk-1 expression was detected with routine ABC method. MATN OUTCOME MEASURES： ① Distribution of pEIk-1 immuno-positive neurons in whole brain of rats in the normal control group. ②Comparison of the expression of pEIk-1 immuno-positive neurons in whole brain at different time points after training between the training group and sham-training group. RESULTS ： All the 55 rats were involved in result analysis. ③ Distribution of pEIk-1 immuno-positive neurons in the whole brain of rats in the normal control group： Strong expressions of pEIk-1 immuno-positive neurons were observed in prefrontal lobe, granular layer of olfactory bulbs, Purkinje cell layer and granular layer of cerebellum, whole stdate cortex, temporal cortex, pre-pyriform cortex, hypothalamic supraoptic nucleus, hypothalamic paraventricular nucleus and periventricular nucleus, thalamic paraventricular nucleus, pronucleus and postnucleus of amygdala cortex, central nucleus of amygdala, medial amygdaloid nucleus, entorhinal cortex, hippocampal dentate gyros, CA1-4 regions, caudate-putamen, material division, brain stem spinal nucleus of trigeminal nerve, and superior olivary nucleus, and those in hippocampal dentate gyrus and CA1 region were the strongest.② Distribution of pEIk-1 immuno-positive neurons in the whole brain of rats at different time points after training in the training group and sham-training group： In the training group, the expressions were obviously enhanced in caudate-putamen of striatum, material division, most cortexes, hippocampal dentate gyrus, hippocampal CA regions, nucleus amygdalae, thalamic paraventricular nucleus, Purkinje cell layer of cerebellum, entorhinal cortex, hypothalamic supraoptic nucleus, hypothalamic paraventricular nucleus, and periventricular nucleus at 0 hour after training, and the enhancement lasted for 6 hours at least, and those at 24 hours were decreased to normal. In the sham-training group, obvious enhanced expressions of pEIk-1 immuno-positive neurons could be observed in most cortexes, nucleus amygdalae, entorhinal cortex, hypothalamic supraoptic nucleus, hypothalamic paraventdoular nucleus and periventricular nucleus, brain stem spinal nucleus of trigeminal nerve, Purkinje cell layer and granular layer of cerebellum at O, 1, 3 and 6 hours, and decreased to normal after 24 hours. The expressions in material division, caudate-putamen of striatum, hippocampus were not obviously enhanced as compared with those in the normal control group, but significantly different from those in the training group （0, 1, 3, 6 hours after training, material division： F= 0.576, 0.023, 0.116, 8.873, P〈 0.01; caudate-putamen： F= 0.157, 0.427, 0.030, 0.001, P〈 0.01; hippocampus： F= 6.716, 2.405, 14.137, 1.416, P 〈 0.05-0.01 ）. CONCLUSION： The expression of activated pEIk-1 can be detected in the learning related brain areas under normal status, and the perk-1 expression in the brain areas dynamically changed in a time-dependent manner after Y-maze training, and it is indicated that pEIk-1 is involved in the learning and memory process in Y-maze related brain areas.展开更多
As deep neural networks (DNNs) have been successfully adopted in various domains, the training of these large-scale models becomes increasingly difficult and is often deployed on compute clusters composed of many devi...As deep neural networks (DNNs) have been successfully adopted in various domains, the training of these large-scale models becomes increasingly difficult and is often deployed on compute clusters composed of many devices like GPUs. However, as the size of the cluster increases, so does the possibility of failures during training. Currently, faults are mainly handled by recording checkpoints and recovering, but this approach causes large overhead and affects the training efficiency even when no error occurs. The low checkpointing frequency leads to a large loss of training time, while the high recording frequency affects the training efficiency. To solve this contradiction, we propose BAFT, a bubble-aware fault tolerant framework for hybrid parallel distributed training. BAFT can automatically analyze parallel strategies, profile the runtime information, and schedule checkpointing tasks at the granularity of pipeline stage depending on the bubble distribution in the training. It supports higher checkpoint efficiency and only introduces less than 1% time overhead, which allows us to record checkpoints at high frequency, thereby reducing the time loss in error recovery and avoiding the impact of fault tolerance on training.展开更多
Edge machine learning creates a new computational paradigm by enabling the deployment of intelligent applications at the network edge.It enhances application efficiency and responsiveness by performing inference and t...Edge machine learning creates a new computational paradigm by enabling the deployment of intelligent applications at the network edge.It enhances application efficiency and responsiveness by performing inference and training tasks closer to data sources.However,it encounters several challenges in practice.The variance in hardware specifications and performance across different devices presents a major issue for the training and inference tasks.Additionally,edge devices typically possess limited network bandwidth and computing resources compared with data centers.Moreover,existing distributed training architectures often fail to consider the constraints of resources and communication efficiency in edge environments.In this paper,we propose DSparse,a method for distributed training based on sparse update in edge clusters with various memory capacities.It aims at maximizing the utilization of memory resources across all devices within a cluster.To reduce memory consumption during the training process,we adopt sparse update to prioritize the updating of selected layers on the devices in the cluster,which not only lowers memory usage but also reduces the data volume of parameters and the time required for parameter aggregation.Furthermore,DSparse utilizes a parameter aggregation mechanism based on multi-process groups,subdividing the aggregation tasks into AllReduce and Broadcast types,thereby further reducing the communication frequency for parameter aggregation.Experimental results using the MobileNetV2 model on the CIFAR-10 dataset demonstrate that DSparse reduces memory consumption by an average of 59.6%across seven devices,with a 75.4%reduction in parameter aggregation time,while maintaining model precision.展开更多
To achieve better performance,researchers have recently focused on building larger deep learning models,substantially increasing the training costs and prompting the development of distributed training within GPU clus...To achieve better performance,researchers have recently focused on building larger deep learning models,substantially increasing the training costs and prompting the development of distributed training within GPU clusters.However,conventional distributed training approaches suffer from limitations:data parallelism is hindered by excessive memory demands and communication overhead during gradient synchronization,while model parallelism fails to achieve optimal device utilization due to strict computational dependencies.To overcome these challenges,researchers have proposed the concept of hybrid parallelism.By segmenting the model into multiple stages that may internally utilize data parallelism and sequentially processing split training data in a pipeline-like manner across different stages,hybrid parallelism enhances model training speed.However,widely used freezing mechanisms in model fine-tuning,namely canceling gradient computation and weight updates for converged parameters to reduce computational overhead,are yet to be efficiently integrated within hybrid parallel training,failing to strike a balance between speeding up training and guaranteeing accuracy and further reducing the time required for the model to reach a converged state.In this paper,we propose Reinforcement Learning Freeze(RLFreeze),a freezing strategy for distributed DNN training in heterogeneous GPU clusters,especially in hybrid parallelism.We first introduce a mixed freezing criterion based on gradients and gradient variation to accurately freeze converged parameters while minimizing the freezing of unconverged ones.Then,RLFreeze selects the parameters to be frozen according to this criterion and dynamically adjusts the required thresholds for freezing decisions during training using reinforcement learning,achieving a balance between accuracy and accelerated model training.Experimental results demonstrate that RLFreeze can further improve training efficiency in both data parallelism and hybrid parallelism while maintaining model accuracy.展开更多
Beamforming is significant for millimeter wave multi-user massive multi-input multi-output systems.In the meanwhile,the overhead cost of channel state information and beam training is considerable,especially in dynami...Beamforming is significant for millimeter wave multi-user massive multi-input multi-output systems.In the meanwhile,the overhead cost of channel state information and beam training is considerable,especially in dynamic environments.To reduce the overhead cost,we propose a multi-user beam tracking algorithm using a distributed deep Q-learning method.With online learning of users’moving trajectories,the proposed algorithm learns to scan a beam subspace to maximize the average effective sum rate.Considering practical implementation,we model the continuous beam tracking problem as a non-Markov decision process and thus develop a simplified training scheme of deep Q-learning to reduce the training complexity.Furthermore,we propose a scalable state-action-reward design for scenarios with different users and antenna numbers.Simulation results verify the effectiveness of the designed method.展开更多
Objective The term ＂pockmark＂ was introduced by King and MacLean （1970） to describe small ＂circular＂ on echosounder records in Nova Scotia. described as circular, near Pockmarks are usually circular or elongated...Objective The term ＂pockmark＂ was introduced by King and MacLean （1970） to describe small ＂circular＂ on echosounder records in Nova Scotia. described as circular, near Pockmarks are usually circular or elongated depressions, generally 10--400 m in diameter and 30-50 m in deep. Pockmarks are normally regarded to be manifestations of fluids escape through the seabed. Pockmarks are valuable features on the seafloor and are useful in constraining the hydrodynamics of sedimentary basins. Since then pockmarks have been recognized in many areas around the world. They occur predominantly in fine-grained siliciclastic depositional settings, although a few case studies have been reported in carbonate settings. In this paper we illustrate a suite of fluid escape features, discovered during the course of petroleum exploration on the West Africa continental margin （Fig. 1）. They are particularly of interest to the oil and gas industry because they could be potential indicators of deeply buried hydrocarbon reservoirs, and fluid flow phenomena in the deep water oilfield are important for the safe and efficient exploration, development and production of hydrocarbons in the area.展开更多
Studies have indicated that the distributed compressed sensing based(DCSbased) channel estimation can decrease the length of the reference signals effectively. In block transmission, a unique word(UW) can be used as a...Studies have indicated that the distributed compressed sensing based(DCSbased) channel estimation can decrease the length of the reference signals effectively. In block transmission, a unique word(UW) can be used as a cyclic prefix and reference signal. However, the DCS-based channel estimation requires diversity sequences instead of UW. In this paper, we proposed a novel method that employs a training sequence(TS) whose duration time is slightly longer than the maximum delay spread time. Based on proposed TS, the DCS approach perform perfectly in multipath channel estimation. Meanwhile, a cyclic prefix construct could be formed, which reduces the complexity of the frequency domain equalization(FDE) directly. Simulation results demonstrate that, by using the method of simultaneous orthogonal matching pursuit(SOMP), the required channel overhead has been reduced thanks to the proposed TS.展开更多
基金the Researchers Supporting Project,King Saud University,Saudi Arabia,for funding this research work through Project No.RSPD2025R951.
文摘This research introduces a unique approach to segmenting breast cancer images using a U-Net-based architecture.However,the computational demand for image processing is very high.Therefore,we have conducted this research to build a system that enables image segmentation training with low-power machines.To accomplish this,all data are divided into several segments,each being trained separately.In the case of prediction,the initial output is predicted from each trained model for an input,where the ultimate output is selected based on the pixel-wise majority voting of the expected outputs,which also ensures data privacy.In addition,this kind of distributed training system allows different computers to be used simultaneously.That is how the training process takes comparatively less time than typical training approaches.Even after completing the training,the proposed prediction system allows a newly trained model to be included in the system.Thus,the prediction is consistently more accurate.We evaluated the effectiveness of the ultimate output based on four performance matrices:average pixel accuracy,mean absolute error,average specificity,and average balanced accuracy.The experimental results show that the scores of average pixel accuracy,mean absolute error,average specificity,and average balanced accuracy are 0.9216,0.0687,0.9477,and 0.8674,respectively.In addition,the proposed method was compared with four other state-of-the-art models in terms of total training time and usage of computational resources.And it outperformed all of them in these aspects.
文摘With the development of economic globalization,distributedmanufacturing is becomingmore andmore prevalent.Recently,integrated scheduling of distributed production and assembly has captured much concern.This research studies a distributed flexible job shop scheduling problem with assembly operations.Firstly,a mixed integer programming model is formulated to minimize the maximum completion time.Secondly,a Q-learning-assisted coevolutionary algorithmis presented to solve themodel:(1)Multiple populations are developed to seek required decisions simultaneously;(2)An encoding and decoding method based on problem features is applied to represent individuals;(3)A hybrid approach of heuristic rules and random methods is employed to acquire a high-quality population;(4)Three evolutionary strategies having crossover and mutation methods are adopted to enhance exploration capabilities;(5)Three neighborhood structures based on problem features are constructed,and a Q-learning-based iterative local search method is devised to improve exploitation abilities.The Q-learning approach is applied to intelligently select better neighborhood structures.Finally,a group of instances is constructed to perform comparison experiments.The effectiveness of the Q-learning approach is verified by comparing the developed algorithm with its variant without the Q-learning method.Three renowned meta-heuristic algorithms are used in comparison with the developed algorithm.The comparison results demonstrate that the designed method exhibits better performance in coping with the formulated problem.
基金supported by the Natural Science Foundation of Anhui Province(Grant Number 2208085MG181)the Science Research Project of Higher Education Institutions in Anhui Province,Philosophy and Social Sciences(Grant Number 2023AH051063)the Open Fund of Key Laboratory of Anhui Higher Education Institutes(Grant Number CS2021-ZD01).
文摘The distributed flexible job shop scheduling problem(DFJSP)has attracted great attention with the growth of the global manufacturing industry.General DFJSP research only considers machine constraints and ignores worker constraints.As one critical factor of production,effective utilization of worker resources can increase productivity.Meanwhile,energy consumption is a growing concern due to the increasingly serious environmental issues.Therefore,the distributed flexible job shop scheduling problem with dual resource constraints(DFJSP-DRC)for minimizing makespan and total energy consumption is studied in this paper.To solve the problem,we present a multi-objective mathematical model for DFJSP-DRC and propose a Q-learning-based multi-objective grey wolf optimizer(Q-MOGWO).In Q-MOGWO,high-quality initial solutions are generated by a hybrid initialization strategy,and an improved active decoding strategy is designed to obtain the scheduling schemes.To further enhance the local search capability and expand the solution space,two wolf predation strategies and three critical factory neighborhood structures based on Q-learning are proposed.These strategies and structures enable Q-MOGWO to explore the solution space more efficiently and thus find better Pareto solutions.The effectiveness of Q-MOGWO in addressing DFJSP-DRC is verified through comparison with four algorithms using 45 instances.The results reveal that Q-MOGWO outperforms comparison algorithms in terms of solution quality.
基金partially supported by the Guangdong Basic and Applied Basic Research Foundation(2023A1515011531)the National Natural Science Foundation of China under Grant 62173356+2 种基金the Science and Technology Development Fund(FDCT),Macao SAR,under Grant 0019/2021/AZhuhai Industry-University-Research Project with Hongkong and Macao under Grant ZH22017002210014PWCthe Key Technologies for Scheduling and Optimization of Complex Distributed Manufacturing Systems(22JR10KA007).
文摘The flow shop scheduling problem is important for the manufacturing industry.Effective flow shop scheduling can bring great benefits to the industry.However,there are few types of research on Distributed Hybrid Flow Shop Problems(DHFSP)by learning assisted meta-heuristics.This work addresses a DHFSP with minimizing the maximum completion time(Makespan).First,a mathematical model is developed for the concerned DHFSP.Second,four Q-learning-assisted meta-heuristics,e.g.,genetic algorithm(GA),artificial bee colony algorithm(ABC),particle swarm optimization(PSO),and differential evolution(DE),are proposed.According to the nature of DHFSP,six local search operations are designed for finding high-quality solutions in local space.Instead of randomselection,Q-learning assists meta-heuristics in choosing the appropriate local search operations during iterations.Finally,based on 60 cases,comprehensive numerical experiments are conducted to assess the effectiveness of the proposed algorithms.The experimental results and discussions prove that using Q-learning to select appropriate local search operations is more effective than the random strategy.To verify the competitiveness of the Q-learning assistedmeta-heuristics,they are compared with the improved iterated greedy algorithm(IIG),which is also for solving DHFSP.The Friedman test is executed on the results by five algorithms.It is concluded that the performance of four Q-learning-assisted meta-heuristics are better than IIG,and the Q-learning-assisted PSO shows the best competitiveness.
基金supported by the National Natural Science Foundation of China(52372310)the State Key Laboratory of Advanced Rail Autonomous Operation(RAO2023ZZ001)+1 种基金the Fundamental Research Funds for the Central Universities(2022JBQY001)Beijing Laboratory of Urban Rail Transit.
文摘The emerging virtual coupling technology aims to operate multiple train units in a Virtually Coupled Train Set(VCTS)at a minimal but safe distance.To guarantee collision avoidance,the safety distance should be calculated using the state-of-the-art space-time separation principle that separates the Emergency Braking(EB)trajectories of two successive units during the whole EB process.In this case,the minimal safety distance is usually numerically calculated without an analytic formulation.Thus,the constrained VCTS control problem is hard to address with space-time separation,which is still a gap in the existing literature.To solve this problem,we propose a Distributed Economic Model Predictive Control(DEMPC)approach with computation efficiency and theoretical guarantee.Specifically,to alleviate the computation burden,we transform implicit safety constraints into explicitly linear ones,such that the optimal control problem in DEMPC is a quadratic programming problem that can be solved efficiently.For theoretical analysis,sufficient conditions are derived to guarantee the recursive feasibility and stability of DEMPC,employing compatibility constraints,tube techniques and terminal ingredient tuning.Moreover,we extend our approach with globally optimal and distributed online EB configuration methods to shorten the minimal distance among VCTS.Finally,experimental results demonstrate the performance and advantages of the proposed approaches.
基金supported by the National Natural Science Foundation of China(No.U21B2003,62072250,62072250,62172435,U1804263,U20B2065,61872203,71802110,61802212)the National Key R&D Program of China(No.2021QY0700)+4 种基金the Key Laboratory of Intelligent Support Technology for Complex Environments(Nanjing University of Information Science and Technology),Ministry of Education,and the Natural Science Foundation of Jiangsu Province(No.BK20200750)Open Foundation of Henan Key Laboratory of Cyberspace Situation Awareness(No.HNTS2022002)Post Graduate Research&Practice Innvoation Program of Jiangsu Province(No.KYCX200974)Open Project Fund of Shandong Provincial Key Laboratory of Computer Network(No.SDKLCN-2022-05)the Priority Academic Program Development of Jiangsu Higher Education Institutions(PAPD)Fund and Graduate Student Scientific Research Innovation Projects of Jiangsu Province(No.KYCX231359).
文摘In recent years,various adversarial defense methods have been proposed to improve the robustness of deep neural networks.Adversarial training is one of the most potent methods to defend against adversarial attacks.However,the difference in the feature space between natural and adversarial examples hinders the accuracy and robustness of the model in adversarial training.This paper proposes a learnable distribution adversarial training method,aiming to construct the same distribution for training data utilizing the Gaussian mixture model.The distribution centroid is built to classify samples and constrain the distribution of the sample features.The natural and adversarial examples are pushed to the same distribution centroid to improve the accuracy and robustness of the model.The proposed method generates adversarial examples to close the distribution gap between the natural and adversarial examples through an attack algorithm explicitly designed for adversarial training.This algorithm gradually increases the accuracy and robustness of the model by scaling perturbation.Finally,the proposed method outputs the predicted labels and the distance between the sample and the distribution centroid.The distribution characteristics of the samples can be utilized to detect adversarial cases that can potentially evade the model defense.The effectiveness of the proposed method is demonstrated through comprehensive experiments.
基金Supported by the National Natural Science Foundation of China(61401496)。
文摘The traditional distributed tactical training simulation system is limited by the availability and bandwidth of military network transmission channel and does not have the feasibility of remote interconnection and spatial sub-regional deployment.In this paper,a new communication method of distributed tactical training simulation system is proposed to solve the problem of strong business coupling between nodes and system availability under the condition of low bandwidth.The operator of federated exchange,federated queue and their concepts and design requirements are firstly proposed,and the inverted tree,triangle,ring exchange topology and circular queue structure are further constructed.Theoretically,the expected goal of high-speed interworking between nodes in the cluster and high reliable transmission between clusters is realized.The example also shows that this method can significantly improve the throughput of single switching node and federated node after using reliability confirmation mechanism.
基金supported by the National Natural Science Foundation of China(U1435220)
文摘How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classification due to the powerful feature representation ability and better performance. However,the training and testing of CNN mainly rely on single machine.Single machine has its natural limitation and bottleneck in processing RSIs due to limited hardware resources and huge time consuming. Besides, overfitting is a challenge for the CNN model due to the unbalance between RSIs data and the model structure.When a model is complex or the training data is relatively small,overfitting occurs and leads to a poor predictive performance. To address these problems, a distributed CNN architecture for RSIs target classification is proposed, which dramatically increases the training speed of CNN and system scalability. It improves the storage ability and processing efficiency of RSIs. Furthermore,Bayesian regularization approach is utilized in order to initialize the weights of the CNN extractor, which increases the robustness and flexibility of the CNN model. It helps prevent the overfitting and avoid the local optima caused by limited RSI training images or the inappropriate CNN structure. In addition, considering the efficiency of the Na¨?ve Bayes classifier, a distributed Na¨?ve Bayes classifier is designed to reduce the training cost. Compared with other algorithms, the proposed system and method perform the best and increase the recognition accuracy. The results show that the distributed system framework and the proposed algorithms are suitable for RSIs target classification tasks.
基金supported by the National Natural Science Foundation of China (Grant No. 12302238)the National Key Research and Development Program of China (Grant Nos. 2021YFB3400701, 2022YFB3402904)。
文摘The dynamic load distribution within in-service axlebox bearings of high-speed trains is crucial for the fatigue reliability assessment and forward design of axlebox bearings. This paper presents an in situ measurement of the dynamic load distribution in the four rows of two axlebox bearings on a bogie wheelset of a high-speed train under polygonal wheel–rail excitation. The measurement employed an improved strain-based method to measure the dynamic radial load distribution of roller bearings. The four rows of two axlebox bearings on a wheelset exhibited different ranges of loaded zones and different means of distributed loads. Besides, the mean value and standard deviation of measured roller–raceway contact loads showed non-monotonic variations with the frequency of wheel–rail excitation. The fatigue life of the four bearing rows under polygonal wheel–rail excitation was quantitatively predicted by compiling the measured roller–raceway contact load spectra of the most loaded position and considering the load spectra as input.
基金Supported by Shandong Provincial Natural Science Foundation of China(Grant No.ZR2022ME180)Guangdong Provincial Basic and Applied Basic Research Fund of China(Grant No.2019A1515111005)National Natural Science Foundation of China(Grant No.51705267)。
文摘The current research on the aerodynamic performance of the train running in rainy weather is primarily concerned with the trajectory of the raindrops and the aerodynamic variation of trains caused by raindrops.In fact,water film will generate on the train body when raindrops hit the train,which interacts with the flow field around the train,and would probably affect the aerodynamic performance of the train.In this paper,based on shear stress transport(SST)k-w turbulence model and Euler-Lagrange discrete phase model,the aerodynamic calculation model of a highspeed train under rainfall environment is established.The LWF(Lagrangian wall film)is used to simulate the water film distribution of the high-speed train under different rainfall intensities,and the aerodynamic performance of the train are studied.The calculation results show that raindrops will gather on the train surface and form water film under rainfall environment.With the extension of rainfall time,the thickness and coverage range of water film expand,and the maximum thickness of water film can reach 4.95 mm under the working conditions in this paper.The average thickness of water film on the train body increases with the rainfall intensity.When the rainfall intensity increases from 100 mm/h to 500 mm/h,the average water film thickness will increase by 3.26 times.The velocity of water film in the streamlined area of head car is larger than that in other areas,and the maximum velocity is 22.14 m/s.Compared with the rainless environment condition,the skin friction coefficient of the high-speed train increases and the average value will increase by 10.74%for a rainfall intensity of 500 mm/h.The positive pressure and resistance coefficient of the head car increase with the rainfall intensity.This research proposes a methodology to systematically analyze the generation of water film on the train surface and its influence on the train aerodynamic performance;the analysis can provide more practical results and can serve as a reference basis for the design and development of high-speed trains.
文摘BACKGROUND： Elk-1 mRNA distributes extensively in the neurons of mice, rat and human brains, and the Elk-1 expression may be correlated with the synaptic plasticity, learning and memory. OBJECTIVE： To observe the distribution of phosphorylated Elk-1 （pEIk-1） in whole brain of rats received Y-maze active avoidance training and the changes of pEIk-1 expression at different time points after training. DESIGN ： A randomized controlled study SETTING ： Research Room of Neurobiology, the Second Affiliated Hospital of Southern Medical University MATERIALS ： Fifty-five male clean-degree SD rats of 3-4 months old, weighing 200-250 g, were provided by the Experimental Animal Center of Southem Medical University. The rabbit anti-monoclonal pEIk-1 antibody was purchased from Cell Signal Transduction Company, and ABC kit from Vector Company. METHODS ： The experiments were carried out in the Research Room of Neurobiology, Second Affiliated Hospital of Southern Medical University from September 2004 to February 2005. ① Grouping： The rats were randomly divided into training group （n = 25）, sham-training group （n = 25） and normal control group （n = 5）, and the training and sham-training groups were observed at 0, 1, 3, 6 and 24 hours after training, which represented the five phases in the process of leaming and memory. ② Y-maze training： The rats were preconditioned in the electrical Y-maze apparatus, 20 minutes a day for 3 days continuously, and training began from the 4^th day. In the training group, the rats were trained with the combination of light and electddty. Each rat repeated for 60 times in each training, and the correct times were recorded, those correct for less than 25 times were taken as unqualified, and excluded from the training group, and supplemented by other rats in time. In the sham-training group, there was no fixed correlation between the application of light and electricity. The rats in the normal contrel group were given not any training. ③Detection of pEIk-1 expression： The rats were anesthetized after Y-maze training, brain tissue was removed to prepare coronal freezing sections, and the pEIk-1 expression was detected with routine ABC method. MATN OUTCOME MEASURES： ① Distribution of pEIk-1 immuno-positive neurons in whole brain of rats in the normal control group. ②Comparison of the expression of pEIk-1 immuno-positive neurons in whole brain at different time points after training between the training group and sham-training group. RESULTS ： All the 55 rats were involved in result analysis. ③ Distribution of pEIk-1 immuno-positive neurons in the whole brain of rats in the normal control group： Strong expressions of pEIk-1 immuno-positive neurons were observed in prefrontal lobe, granular layer of olfactory bulbs, Purkinje cell layer and granular layer of cerebellum, whole stdate cortex, temporal cortex, pre-pyriform cortex, hypothalamic supraoptic nucleus, hypothalamic paraventricular nucleus and periventricular nucleus, thalamic paraventricular nucleus, pronucleus and postnucleus of amygdala cortex, central nucleus of amygdala, medial amygdaloid nucleus, entorhinal cortex, hippocampal dentate gyros, CA1-4 regions, caudate-putamen, material division, brain stem spinal nucleus of trigeminal nerve, and superior olivary nucleus, and those in hippocampal dentate gyrus and CA1 region were the strongest.② Distribution of pEIk-1 immuno-positive neurons in the whole brain of rats at different time points after training in the training group and sham-training group： In the training group, the expressions were obviously enhanced in caudate-putamen of striatum, material division, most cortexes, hippocampal dentate gyrus, hippocampal CA regions, nucleus amygdalae, thalamic paraventricular nucleus, Purkinje cell layer of cerebellum, entorhinal cortex, hypothalamic supraoptic nucleus, hypothalamic paraventricular nucleus, and periventricular nucleus at 0 hour after training, and the enhancement lasted for 6 hours at least, and those at 24 hours were decreased to normal. In the sham-training group, obvious enhanced expressions of pEIk-1 immuno-positive neurons could be observed in most cortexes, nucleus amygdalae, entorhinal cortex, hypothalamic supraoptic nucleus, hypothalamic paraventdoular nucleus and periventricular nucleus, brain stem spinal nucleus of trigeminal nerve, Purkinje cell layer and granular layer of cerebellum at O, 1, 3 and 6 hours, and decreased to normal after 24 hours. The expressions in material division, caudate-putamen of striatum, hippocampus were not obviously enhanced as compared with those in the normal control group, but significantly different from those in the training group （0, 1, 3, 6 hours after training, material division： F= 0.576, 0.023, 0.116, 8.873, P〈 0.01; caudate-putamen： F= 0.157, 0.427, 0.030, 0.001, P〈 0.01; hippocampus： F= 6.716, 2.405, 14.137, 1.416, P 〈 0.05-0.01 ）. CONCLUSION： The expression of activated pEIk-1 can be detected in the learning related brain areas under normal status, and the perk-1 expression in the brain areas dynamically changed in a time-dependent manner after Y-maze training, and it is indicated that pEIk-1 is involved in the learning and memory process in Y-maze related brain areas.
基金supported by the National Key R&D Program of China(2021ZD0110104)the National Natural Science Foundation of China(Grant Nos.62222210,U21B2017,61832006,and 62072297).
文摘As deep neural networks (DNNs) have been successfully adopted in various domains, the training of these large-scale models becomes increasingly difficult and is often deployed on compute clusters composed of many devices like GPUs. However, as the size of the cluster increases, so does the possibility of failures during training. Currently, faults are mainly handled by recording checkpoints and recovering, but this approach causes large overhead and affects the training efficiency even when no error occurs. The low checkpointing frequency leads to a large loss of training time, while the high recording frequency affects the training efficiency. To solve this contradiction, we propose BAFT, a bubble-aware fault tolerant framework for hybrid parallel distributed training. BAFT can automatically analyze parallel strategies, profile the runtime information, and schedule checkpointing tasks at the granularity of pipeline stage depending on the bubble distribution in the training. It supports higher checkpoint efficiency and only introduces less than 1% time overhead, which allows us to record checkpoints at high frequency, thereby reducing the time loss in error recovery and avoiding the impact of fault tolerance on training.
基金supported by the National Natural Science Foundation of China under Grant Nos.62072434 and U23B2004the Innovation Funding of Institute of Computing Technology,Chinese Academy of Sciences,under Grant Nos.E361050 and E361030.
文摘Edge machine learning creates a new computational paradigm by enabling the deployment of intelligent applications at the network edge.It enhances application efficiency and responsiveness by performing inference and training tasks closer to data sources.However,it encounters several challenges in practice.The variance in hardware specifications and performance across different devices presents a major issue for the training and inference tasks.Additionally,edge devices typically possess limited network bandwidth and computing resources compared with data centers.Moreover,existing distributed training architectures often fail to consider the constraints of resources and communication efficiency in edge environments.In this paper,we propose DSparse,a method for distributed training based on sparse update in edge clusters with various memory capacities.It aims at maximizing the utilization of memory resources across all devices within a cluster.To reduce memory consumption during the training process,we adopt sparse update to prioritize the updating of selected layers on the devices in the cluster,which not only lowers memory usage but also reduces the data volume of parameters and the time required for parameter aggregation.Furthermore,DSparse utilizes a parameter aggregation mechanism based on multi-process groups,subdividing the aggregation tasks into AllReduce and Broadcast types,thereby further reducing the communication frequency for parameter aggregation.Experimental results using the MobileNetV2 model on the CIFAR-10 dataset demonstrate that DSparse reduces memory consumption by an average of 59.6%across seven devices,with a 75.4%reduction in parameter aggregation time,while maintaining model precision.
基金supported by the Science and Technology project of State Grid Jiangsu Electric Power CO.LTD.under Grants No.J2023153.
文摘To achieve better performance,researchers have recently focused on building larger deep learning models,substantially increasing the training costs and prompting the development of distributed training within GPU clusters.However,conventional distributed training approaches suffer from limitations:data parallelism is hindered by excessive memory demands and communication overhead during gradient synchronization,while model parallelism fails to achieve optimal device utilization due to strict computational dependencies.To overcome these challenges,researchers have proposed the concept of hybrid parallelism.By segmenting the model into multiple stages that may internally utilize data parallelism and sequentially processing split training data in a pipeline-like manner across different stages,hybrid parallelism enhances model training speed.However,widely used freezing mechanisms in model fine-tuning,namely canceling gradient computation and weight updates for converged parameters to reduce computational overhead,are yet to be efficiently integrated within hybrid parallel training,failing to strike a balance between speeding up training and guaranteeing accuracy and further reducing the time required for the model to reach a converged state.In this paper,we propose Reinforcement Learning Freeze(RLFreeze),a freezing strategy for distributed DNN training in heterogeneous GPU clusters,especially in hybrid parallelism.We first introduce a mixed freezing criterion based on gradients and gradient variation to accurately freeze converged parameters while minimizing the freezing of unconverged ones.Then,RLFreeze selects the parameters to be frozen according to this criterion and dynamically adjusts the required thresholds for freezing decisions during training using reinforcement learning,achieving a balance between accuracy and accelerated model training.Experimental results demonstrate that RLFreeze can further improve training efficiency in both data parallelism and hybrid parallelism while maintaining model accuracy.
文摘Beamforming is significant for millimeter wave multi-user massive multi-input multi-output systems.In the meanwhile,the overhead cost of channel state information and beam training is considerable,especially in dynamic environments.To reduce the overhead cost,we propose a multi-user beam tracking algorithm using a distributed deep Q-learning method.With online learning of users’moving trajectories,the proposed algorithm learns to scan a beam subspace to maximize the average effective sum rate.Considering practical implementation,we model the continuous beam tracking problem as a non-Markov decision process and thus develop a simplified training scheme of deep Q-learning to reduce the training complexity.Furthermore,we propose a scalable state-action-reward design for scenarios with different users and antenna numbers.Simulation results verify the effectiveness of the designed method.
基金supported by the National Planned Major Science and Technology Projects of China(grant No.2011ZX05030-005-02)
文摘Objective The term ＂pockmark＂ was introduced by King and MacLean （1970） to describe small ＂circular＂ on echosounder records in Nova Scotia. described as circular, near Pockmarks are usually circular or elongated depressions, generally 10--400 m in diameter and 30-50 m in deep. Pockmarks are normally regarded to be manifestations of fluids escape through the seabed. Pockmarks are valuable features on the seafloor and are useful in constraining the hydrodynamics of sedimentary basins. Since then pockmarks have been recognized in many areas around the world. They occur predominantly in fine-grained siliciclastic depositional settings, although a few case studies have been reported in carbonate settings. In this paper we illustrate a suite of fluid escape features, discovered during the course of petroleum exploration on the West Africa continental margin （Fig. 1）. They are particularly of interest to the oil and gas industry because they could be potential indicators of deeply buried hydrocarbon reservoirs, and fluid flow phenomena in the deep water oilfield are important for the safe and efficient exploration, development and production of hydrocarbons in the area.
基金support by National Key Technology Research and Development Program of the Ministry of Science and Technology of China (2015BAK05B01)
文摘Studies have indicated that the distributed compressed sensing based(DCSbased) channel estimation can decrease the length of the reference signals effectively. In block transmission, a unique word(UW) can be used as a cyclic prefix and reference signal. However, the DCS-based channel estimation requires diversity sequences instead of UW. In this paper, we proposed a novel method that employs a training sequence(TS) whose duration time is slightly longer than the maximum delay spread time. Based on proposed TS, the DCS approach perform perfectly in multipath channel estimation. Meanwhile, a cyclic prefix construct could be formed, which reduces the complexity of the frequency domain equalization(FDE) directly. Simulation results demonstrate that, by using the method of simultaneous orthogonal matching pursuit(SOMP), the required channel overhead has been reduced thanks to the proposed TS.
