In this paper, a new reliable hierarchical model is suggested for a two-wagon train Networked Control System. Each wagon has a Controller that carries the control load and an Entertainment server that handles the ente...In this paper, a new reliable hierarchical model is suggested for a two-wagon train Networked Control System. Each wagon has a Controller that carries the control load and an Entertainment server that handles the entertainment. A supervisory controller runs on top of the two controllers and the two entertainment servers. Contrary to a similar model in the literature, the Supervisory node replaces a Controller as soon as it fails (Active Supervisor). All system states are analyzed and simulated using OPNET. It is shown that, for all states, this architecture has zero control packets dropped and the end-to-end delay is below the maximum target delay. A comparison between this Active model and the other model in the literature is presented. It is found that the entertainment in this new architecture is kept available for the passengers in more of the system states when compared to the architecture previously presented in the literature.展开更多
In this paper, a novel reconfiguration technique is developed in the context of a fault-tolerant Networked Control System (NCS) in two train wagons. All sensors, controllers and actuators in both wagons are connected ...In this paper, a novel reconfiguration technique is developed in the context of a fault-tolerant Networked Control System (NCS) in two train wagons. All sensors, controllers and actuators in both wagons are connected on top of a single Gigabit Ethernet network. The network also carries wired and wireless entertainment loads. A Markov model is used to prove that this reconfiguration technique reduces the effect of a failure in the error detection and switching mechanisms on the reliability of the control function. All calculations are based on closed-form solutions and verified using the SHARPE software package.展开更多
Although train modeling research is vast, most available simulation tools are confined to city-or trip-scale analysis, primarily offering micro-level simulations of network segments. This paper addresses this void by ...Although train modeling research is vast, most available simulation tools are confined to city-or trip-scale analysis, primarily offering micro-level simulations of network segments. This paper addresses this void by developing the Ne Train Sim simulator for heavy long-haul freight trains on a network of multiple intersecting tracks. The main objective of this simulator is to enable a comprehensive analysis of energy consumption and the associated carbon footprint for the entire train system. Four case studies were conducted to demonstrate the simulator's performance. The first case study validates the model by comparing Ne Train Sim output to empirical trajectory data. The results demonstrate that the simulated trajectory is precise enough to estimate the train energy consumption and carbon dioxide emissions. The second application demonstrates the train-following model considering six trains following each other. The results showcase the model ability to maintain safefollowing distances between successive trains. The next study highlights the simulator's ability to resolve train conflicts for different scenarios. Finally, the suitability of the Ne Train Sim for modeling realistic railroad networks is verified through the modeling of the entire US network and comparing alternative powertrains on the fleet energy consumption.展开更多
Train network system is a distributed bus system based on MVB bus, and its core component is vehicle control unit (VCU). The software architecture of VCU is developed in a flat and distributed way based on function or...Train network system is a distributed bus system based on MVB bus, and its core component is vehicle control unit (VCU). The software architecture of VCU is developed in a flat and distributed way based on function orientation, which is convenient for expansion, upgrading and testing.展开更多
Many applications of principal component analysis (PCA) can be found in dimensionality reduction. But linear PCA method is not well suitable for nonlinear chemical processes. A new PCA method based on im-proved input ...Many applications of principal component analysis (PCA) can be found in dimensionality reduction. But linear PCA method is not well suitable for nonlinear chemical processes. A new PCA method based on im-proved input training neural network (IT-NN) is proposed for the nonlinear system modelling in this paper. Mo-mentum factor and adaptive learning rate are introduced into learning algorithm to improve the training speed of IT-NN. Contrasting to the auto-associative neural network (ANN), IT-NN has less hidden layers and higher training speed. The effectiveness is illustrated through a comparison of IT-NN with linear PCA and ANN with experiments. Moreover, the IT-NN is combined with RBF neural network (RBF-NN) to model the yields of ethylene and propyl-ene in the naphtha pyrolysis system. From the illustrative example and practical application, IT-NN combined with RBF-NN is an effective method of nonlinear chemical process modelling.展开更多
To handle the handover challenge in Express Train Access Networks(ETAN).mobility fading effects in high speed railway environments should be addressed first.Based on the investigation of fading effects in this paper,w...To handle the handover challenge in Express Train Access Networks(ETAN).mobility fading effects in high speed railway environments should be addressed first.Based on the investigation of fading effects in this paper,we obtain two theoretical bounds:HOTiming upper bound and HO-Margin lower bound,which are helpful guidelines to study the handover challenge today and in the future.Then,we apply them to analyze performance of conventional handover technologies and our proposal in ETAN.This follow-up theory analyses and simulation experiment results demonstrate that the proposed handover solution can minimize handover time up to 4ms(which is the fastest one so far),and reduce HO-Margin to 0.16 dB at a train speed of 350km/h.展开更多
This paper is focused on the technique for de si gn and realization of the process communications about the computer-aided train diagram network system. The Windows Socket technique is adopted to program for the cli...This paper is focused on the technique for de si gn and realization of the process communications about the computer-aided train diagram network system. The Windows Socket technique is adopted to program for the client and the server to create system applications and solve the problems o f data transfer and data sharing in the system.展开更多
The on-board diagnosis network is the nervous system of high-speed Maglev trains, connecting all controller sensors, and corresponding devices to realize the information acquisition and control. In order to study the ...The on-board diagnosis network is the nervous system of high-speed Maglev trains, connecting all controller sensors, and corresponding devices to realize the information acquisition and control. In order to study the on-board diagnosis network＇s security and reliability, a simulation model for the on-board diagnosis network of high-speed Maglev trains with the optimal network engineering tool （OPNET） was built to analyze the network＇s performance, such as response error and bit error rate on the network load, throughput, and node-state response. The simulation model was verified with an actual on-board diagnosis network structure. The results show that the model results obtained are in good agreement with actual system performance and can be used to achieve actual communication network optimization and control algorithms.展开更多
Training deep neural networks(DNNs)requires a significant amount of time and resources to obtain acceptable results,which severely limits its deployment in resource-limited platforms.This paper proposes DarkFPGA,a nov...Training deep neural networks(DNNs)requires a significant amount of time and resources to obtain acceptable results,which severely limits its deployment in resource-limited platforms.This paper proposes DarkFPGA,a novel customizable framework to efficiently accelerate the entire DNN training on a single FPGA platform.First,we explore batch-level parallelism to enable efficient FPGA-based DNN training.Second,we devise a novel hardware architecture optimised by a batch-oriented data pattern and tiling techniques to effectively exploit parallelism.Moreover,an analytical model is developed to determine the optimal design parameters for the DarkFPGA accelerator with respect to a specific network specification and FPGA resource constraints.Our results show that the accelerator is able to perform about 10 times faster than CPU training and about a third of the energy consumption than GPU training using 8-bit integers for training VGG-like networks on the CIFAR dataset for the Maxeler MAX5 platform.展开更多
The existing computer and network technology makes the enterprise training transform from the traditional mode into new mode. The paper studies how to combine enterprise knowledge management and network training to ma...The existing computer and network technology makes the enterprise training transform from the traditional mode into new mode. The paper studies how to combine enterprise knowledge management and network training to make the enterprise training meet the demands of knowledge management and improve the competitiveness of enterprises. And the paper puts forwards the new opinion combining enterprise knowledge management and network training system. The purpose of applying knowledge map and knowledge push to training system is to integrate knowledge management into training system to make the enterprises face the challenge of knowledge economy.展开更多
Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new meth...Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new method based on virtual model and invariant moments was proposed to generate training samples.The method was composed of the following steps：use computer and simulation software to build target object＇s virtual model and then simulate the environment,light condition,camera parameter,etc.;rotate the model by spin and nutation of inclination to get the image sequence by virtual camera;preprocess each image and transfer them into binary image;calculate the invariant moments for each image and get a vectors＇ sequence.The vectors＇ sequence which was proved to be complete became the training samples together with the target outputs.The simulated results showed that the proposed method could be used to recognize the real targets and improve the accuracy of target recognition effectively when the sampling interval was short enough and the circumstance simulation was close enough.展开更多
BACKGROUND Prior studies indicate that doing breathing exercises improves physical performance and quality of life (QoL) in heart failure patients. However, these effects remain unclear and contradictory. AIM To deter...BACKGROUND Prior studies indicate that doing breathing exercises improves physical performance and quality of life (QoL) in heart failure patients. However, these effects remain unclear and contradictory. AIM To determine the effects of machine-assisted and non-machine-assisted respiratory training on physical performance and QoL in heart failure patients. METHODS This was a systematic review and network meta-analysis study. A literature search of electronic databases was conducted for randomized controlled trials (RCTs) on heart failure. Respiratory training interventions were grouped as seven categories: IMT_Pn (inspiratory muscle training without pressure or < 10% maximal inspiratory pressure, MIP), IMT_Pl (inspiratory muscle training with low pressure, 10%-15% MIP), IMT_Pm (inspiratory muscle training with medium pressure, 30%-40% MIP), IMT_Ph (inspiratory muscle training with high pressure, 60% MIP or MIP plus aerobics), Aerobics (aerobic exercise or weight training), Qi_Ex (tai chi, yoga, and breathing exercise), and none. The four outcomes were heart rate, peak oxygen uptake (VO2 peak), 6-min walking distance test (6MWT), and Minnesota Living with Heart Failure QoL. The random-effects model, side-splitting model, and the surface under the cumulative ranking curve (SUCRA) were used to test and analyze the data. RESULTS A total of 1499 subjects from 31 RCT studies were included. IMT_Ph had the highest effect sizes for VO2 peak and 6MWT, IMT_Pm highest for QoL, and Qi_Ex highest for heart rate. Aerobics had the second highest for VO2 peak, Qi_Ex second highest for 6MWT, and IMT_Ph second highest for heart rate and QoL.CONCLUSION This study supports that high- and medium-intensity machine-assisted training improves exercise capacity and QoL in hospital-based heart failure patients. After hospital discharge, non-machine-assisted training continuously improves cardiac function.展开更多
Based on the"Understandable Output Hypothesis"a practical study on the construction of college oral English network training camp is set up,through speech learning and imitation,building language input in na...Based on the"Understandable Output Hypothesis"a practical study on the construction of college oral English network training camp is set up,through speech learning and imitation,building language input in natural environment,exploring effective output mode based on information technology platform,providing foreign language learners with opportunities to express language and get feedback.Students use relevant resources on the Internet to complete the oral activities of"thematic activities"together,so as to cultivate students'cooperative learning,communication skills,team spirit and language communication ability.展开更多
Simulating large-scale and complex systems is commonly considered a difficult and time-consuming task. In this paper, we propose a partial simulation way to speed up the simulation with real time demands. It is based ...Simulating large-scale and complex systems is commonly considered a difficult and time-consuming task. In this paper, we propose a partial simulation way to speed up the simulation with real time demands. It is based on the idea that a train traffic diagram is expressed in a network, and through calculating the maximal long path in the network the simulation is done, but only within a particular partial area.Upon this, we let it become a problem oriented simulation. The simulation could be started at any time,from any trains or at any stations and stopped as the same way according to the problem to be concerned.We can use this kind of simulation to analyse or confirm the correctness of traffic schedule at a high speed to meet the real time demands.展开更多
The high performance of the state-of-the-art deep neural networks(DNNs)is acquired at the cost of huge consumption of computing resources.Quantization of networks is recently recognized as a promising solution to solv...The high performance of the state-of-the-art deep neural networks(DNNs)is acquired at the cost of huge consumption of computing resources.Quantization of networks is recently recognized as a promising solution to solve the problem and significantly reduce the resource usage.However,the previous quantization works have mostly focused on the DNN inference,and there were very few works to address on the challenges of DNN training.In this paper,we leverage dynamic fixed-point(DFP)quantization algorithm and stochastic rounding(SR)strategy to develop a fully quantized 8-bit neural networks targeting low bitwidth training.The experiments show that,in comparison to the full-precision networks,the accuracy drop of our quantized convolutional neural networks(CNNs)can be less than 2%,even when applied to deep models evaluated on Image-Net dataset.Additionally,our 8-bit GNMT translation network can achieve almost identical BLEU to full-precision network.We further implement a prototype on FPGA and the synthesis shows that the low bitwidth training scheme can reduce the resource usage significantly.展开更多
Overfitting is one of the important problems that restrain the application of neural network. The traditional OBD (Optimal Brain Damage) algorithm can avoid overfitting effectively. But it needs to train the network r...Overfitting is one of the important problems that restrain the application of neural network. The traditional OBD (Optimal Brain Damage) algorithm can avoid overfitting effectively. But it needs to train the network repeatedly with low calculational efficiency. In this paper, the Marquardt algorithm is incorporated into the OBD algorithm and a new method for pruning network-the Dynamic Optimal Brain Damage (DOBD) is introduced. This algorithm simplifies a network and obtains good generalization through dynamically deleting weight parameters with low sensitivity that is defined as the change of error function value with respect to the change of weights. Also a simplified method is presented through which sensitivities can be calculated during training with a little computation. A rule to determine the lower limit of sensitivity for deleting the unnecessary weights and other control methods during pruning and training are introduced. The training course is analyzed theoretically and the reason why DOBD algorithm can obtain a much faster training speed than the OBD algorithm and avoid overfitting effectively is given.展开更多
In the first part of the article, a new algorithm for pruning networkDynamic Optimal Brain Damage(DOBD) is introduced. In this part, two cases and an industrial application are worked out to test the new algorithm. It...In the first part of the article, a new algorithm for pruning networkDynamic Optimal Brain Damage(DOBD) is introduced. In this part, two cases and an industrial application are worked out to test the new algorithm. It is verified that the algorithm can obtain good generalization through deleting weight parameters with low sensitivities dynamically and get better result than the Marquardt algorithm or the cross-validation method. Although the initial construction of network may be different, the finial number of free weights pruned by the DOBD algorithm is similar and the number is just close to the optimal number of free weights. The algorithm is also helpful to design the optimal structure of network.展开更多
文摘In this paper, a new reliable hierarchical model is suggested for a two-wagon train Networked Control System. Each wagon has a Controller that carries the control load and an Entertainment server that handles the entertainment. A supervisory controller runs on top of the two controllers and the two entertainment servers. Contrary to a similar model in the literature, the Supervisory node replaces a Controller as soon as it fails (Active Supervisor). All system states are analyzed and simulated using OPNET. It is shown that, for all states, this architecture has zero control packets dropped and the end-to-end delay is below the maximum target delay. A comparison between this Active model and the other model in the literature is presented. It is found that the entertainment in this new architecture is kept available for the passengers in more of the system states when compared to the architecture previously presented in the literature.
文摘In this paper, a novel reconfiguration technique is developed in the context of a fault-tolerant Networked Control System (NCS) in two train wagons. All sensors, controllers and actuators in both wagons are connected on top of a single Gigabit Ethernet network. The network also carries wired and wireless entertainment loads. A Markov model is used to prove that this reconfiguration technique reduces the effect of a failure in the error detection and switching mechanisms on the reliability of the control function. All calculations are based on closed-form solutions and verified using the SHARPE software package.
基金funded in part by the Advanced Research Projects AgencyEnergy (ARPA-E), U.S. Department of Energy, under award number DE-AR0001471。
文摘Although train modeling research is vast, most available simulation tools are confined to city-or trip-scale analysis, primarily offering micro-level simulations of network segments. This paper addresses this void by developing the Ne Train Sim simulator for heavy long-haul freight trains on a network of multiple intersecting tracks. The main objective of this simulator is to enable a comprehensive analysis of energy consumption and the associated carbon footprint for the entire train system. Four case studies were conducted to demonstrate the simulator's performance. The first case study validates the model by comparing Ne Train Sim output to empirical trajectory data. The results demonstrate that the simulated trajectory is precise enough to estimate the train energy consumption and carbon dioxide emissions. The second application demonstrates the train-following model considering six trains following each other. The results showcase the model ability to maintain safefollowing distances between successive trains. The next study highlights the simulator's ability to resolve train conflicts for different scenarios. Finally, the suitability of the Ne Train Sim for modeling realistic railroad networks is verified through the modeling of the entire US network and comparing alternative powertrains on the fleet energy consumption.
文摘Train network system is a distributed bus system based on MVB bus, and its core component is vehicle control unit (VCU). The software architecture of VCU is developed in a flat and distributed way based on function orientation, which is convenient for expansion, upgrading and testing.
基金Supported by Beijing Municipal Education Commission (No.xk100100435) and the Key Research Project of Science andTechnology from Sinopec (No.E03007).
文摘Many applications of principal component analysis (PCA) can be found in dimensionality reduction. But linear PCA method is not well suitable for nonlinear chemical processes. A new PCA method based on im-proved input training neural network (IT-NN) is proposed for the nonlinear system modelling in this paper. Mo-mentum factor and adaptive learning rate are introduced into learning algorithm to improve the training speed of IT-NN. Contrasting to the auto-associative neural network (ANN), IT-NN has less hidden layers and higher training speed. The effectiveness is illustrated through a comparison of IT-NN with linear PCA and ANN with experiments. Moreover, the IT-NN is combined with RBF neural network (RBF-NN) to model the yields of ethylene and propyl-ene in the naphtha pyrolysis system. From the illustrative example and practical application, IT-NN combined with RBF-NN is an effective method of nonlinear chemical process modelling.
基金supported by the National Basic Research Program of China (973 Program)(No.2012CB315606 and 2010CB328201)
文摘To handle the handover challenge in Express Train Access Networks(ETAN).mobility fading effects in high speed railway environments should be addressed first.Based on the investigation of fading effects in this paper,we obtain two theoretical bounds:HOTiming upper bound and HO-Margin lower bound,which are helpful guidelines to study the handover challenge today and in the future.Then,we apply them to analyze performance of conventional handover technologies and our proposal in ETAN.This follow-up theory analyses and simulation experiment results demonstrate that the proposed handover solution can minimize handover time up to 4ms(which is the fastest one so far),and reduce HO-Margin to 0.16 dB at a train speed of 350km/h.
文摘This paper is focused on the technique for de si gn and realization of the process communications about the computer-aided train diagram network system. The Windows Socket technique is adopted to program for the client and the server to create system applications and solve the problems o f data transfer and data sharing in the system.
基金supported by the National Natural Science Foundation of China (No. 51007074)the Program for New Century Excellent Talents in University(NECT-08-0825)+1 种基金the Research and Development Project of the National Railway Ministry (2011J016-B)The basic research universities special fund operations(SWJTU11CX141)
文摘The on-board diagnosis network is the nervous system of high-speed Maglev trains, connecting all controller sensors, and corresponding devices to realize the information acquisition and control. In order to study the on-board diagnosis network＇s security and reliability, a simulation model for the on-board diagnosis network of high-speed Maglev trains with the optimal network engineering tool （OPNET） was built to analyze the network＇s performance, such as response error and bit error rate on the network load, throughput, and node-state response. The simulation model was verified with an actual on-board diagnosis network structure. The results show that the model results obtained are in good agreement with actual system performance and can be used to achieve actual communication network optimization and control algorithms.
文摘Training deep neural networks(DNNs)requires a significant amount of time and resources to obtain acceptable results,which severely limits its deployment in resource-limited platforms.This paper proposes DarkFPGA,a novel customizable framework to efficiently accelerate the entire DNN training on a single FPGA platform.First,we explore batch-level parallelism to enable efficient FPGA-based DNN training.Second,we devise a novel hardware architecture optimised by a batch-oriented data pattern and tiling techniques to effectively exploit parallelism.Moreover,an analytical model is developed to determine the optimal design parameters for the DarkFPGA accelerator with respect to a specific network specification and FPGA resource constraints.Our results show that the accelerator is able to perform about 10 times faster than CPU training and about a third of the energy consumption than GPU training using 8-bit integers for training VGG-like networks on the CIFAR dataset for the Maxeler MAX5 platform.
文摘The existing computer and network technology makes the enterprise training transform from the traditional mode into new mode. The paper studies how to combine enterprise knowledge management and network training to make the enterprise training meet the demands of knowledge management and improve the competitiveness of enterprises. And the paper puts forwards the new opinion combining enterprise knowledge management and network training system. The purpose of applying knowledge map and knowledge push to training system is to integrate knowledge management into training system to make the enterprises face the challenge of knowledge economy.
基金Supported by the Ministerial Level Research Foundation(404040401)
文摘Training neural network to recognize targets needs a lot of samples.People usually get these samples in a non-systematic way,which can miss or overemphasize some target information.To improve this situation,a new method based on virtual model and invariant moments was proposed to generate training samples.The method was composed of the following steps：use computer and simulation software to build target object＇s virtual model and then simulate the environment,light condition,camera parameter,etc.;rotate the model by spin and nutation of inclination to get the image sequence by virtual camera;preprocess each image and transfer them into binary image;calculate the invariant moments for each image and get a vectors＇ sequence.The vectors＇ sequence which was proved to be complete became the training samples together with the target outputs.The simulated results showed that the proposed method could be used to recognize the real targets and improve the accuracy of target recognition effectively when the sampling interval was short enough and the circumstance simulation was close enough.
文摘BACKGROUND Prior studies indicate that doing breathing exercises improves physical performance and quality of life (QoL) in heart failure patients. However, these effects remain unclear and contradictory. AIM To determine the effects of machine-assisted and non-machine-assisted respiratory training on physical performance and QoL in heart failure patients. METHODS This was a systematic review and network meta-analysis study. A literature search of electronic databases was conducted for randomized controlled trials (RCTs) on heart failure. Respiratory training interventions were grouped as seven categories: IMT_Pn (inspiratory muscle training without pressure or < 10% maximal inspiratory pressure, MIP), IMT_Pl (inspiratory muscle training with low pressure, 10%-15% MIP), IMT_Pm (inspiratory muscle training with medium pressure, 30%-40% MIP), IMT_Ph (inspiratory muscle training with high pressure, 60% MIP or MIP plus aerobics), Aerobics (aerobic exercise or weight training), Qi_Ex (tai chi, yoga, and breathing exercise), and none. The four outcomes were heart rate, peak oxygen uptake (VO2 peak), 6-min walking distance test (6MWT), and Minnesota Living with Heart Failure QoL. The random-effects model, side-splitting model, and the surface under the cumulative ranking curve (SUCRA) were used to test and analyze the data. RESULTS A total of 1499 subjects from 31 RCT studies were included. IMT_Ph had the highest effect sizes for VO2 peak and 6MWT, IMT_Pm highest for QoL, and Qi_Ex highest for heart rate. Aerobics had the second highest for VO2 peak, Qi_Ex second highest for 6MWT, and IMT_Ph second highest for heart rate and QoL.CONCLUSION This study supports that high- and medium-intensity machine-assisted training improves exercise capacity and QoL in hospital-based heart failure patients. After hospital discharge, non-machine-assisted training continuously improves cardiac function.
文摘Based on the"Understandable Output Hypothesis"a practical study on the construction of college oral English network training camp is set up,through speech learning and imitation,building language input in natural environment,exploring effective output mode based on information technology platform,providing foreign language learners with opportunities to express language and get feedback.Students use relevant resources on the Internet to complete the oral activities of"thematic activities"together,so as to cultivate students'cooperative learning,communication skills,team spirit and language communication ability.
文摘Simulating large-scale and complex systems is commonly considered a difficult and time-consuming task. In this paper, we propose a partial simulation way to speed up the simulation with real time demands. It is based on the idea that a train traffic diagram is expressed in a network, and through calculating the maximal long path in the network the simulation is done, but only within a particular partial area.Upon this, we let it become a problem oriented simulation. The simulation could be started at any time,from any trains or at any stations and stopped as the same way according to the problem to be concerned.We can use this kind of simulation to analyse or confirm the correctness of traffic schedule at a high speed to meet the real time demands.
文摘The high performance of the state-of-the-art deep neural networks(DNNs)is acquired at the cost of huge consumption of computing resources.Quantization of networks is recently recognized as a promising solution to solve the problem and significantly reduce the resource usage.However,the previous quantization works have mostly focused on the DNN inference,and there were very few works to address on the challenges of DNN training.In this paper,we leverage dynamic fixed-point(DFP)quantization algorithm and stochastic rounding(SR)strategy to develop a fully quantized 8-bit neural networks targeting low bitwidth training.The experiments show that,in comparison to the full-precision networks,the accuracy drop of our quantized convolutional neural networks(CNNs)can be less than 2%,even when applied to deep models evaluated on Image-Net dataset.Additionally,our 8-bit GNMT translation network can achieve almost identical BLEU to full-precision network.We further implement a prototype on FPGA and the synthesis shows that the low bitwidth training scheme can reduce the resource usage significantly.
文摘Overfitting is one of the important problems that restrain the application of neural network. The traditional OBD (Optimal Brain Damage) algorithm can avoid overfitting effectively. But it needs to train the network repeatedly with low calculational efficiency. In this paper, the Marquardt algorithm is incorporated into the OBD algorithm and a new method for pruning network-the Dynamic Optimal Brain Damage (DOBD) is introduced. This algorithm simplifies a network and obtains good generalization through dynamically deleting weight parameters with low sensitivity that is defined as the change of error function value with respect to the change of weights. Also a simplified method is presented through which sensitivities can be calculated during training with a little computation. A rule to determine the lower limit of sensitivity for deleting the unnecessary weights and other control methods during pruning and training are introduced. The training course is analyzed theoretically and the reason why DOBD algorithm can obtain a much faster training speed than the OBD algorithm and avoid overfitting effectively is given.
文摘In the first part of the article, a new algorithm for pruning networkDynamic Optimal Brain Damage(DOBD) is introduced. In this part, two cases and an industrial application are worked out to test the new algorithm. It is verified that the algorithm can obtain good generalization through deleting weight parameters with low sensitivities dynamically and get better result than the Marquardt algorithm or the cross-validation method. Although the initial construction of network may be different, the finial number of free weights pruned by the DOBD algorithm is similar and the number is just close to the optimal number of free weights. The algorithm is also helpful to design the optimal structure of network.
