作者:
Ma, ZhiyiPeking Univ
Sch Elect Engn & Comp Sci Software Inst Beijing Peoples R China Peking Univ
Minist Educ Key Lab High Confidence Software Technol Beijing Peoples R China
There are a large number of modeling languages based on metamodels, and many of the languages are large and complex. In many cases, only part of a metamodel is needed. Hence, it is necessary to automatically extract n...
详细信息
ISBN:
(纸本)9789897582103
There are a large number of modeling languages based on metamodels, and many of the languages are large and complex. In many cases, only part of a metamodel is needed. Hence, it is necessary to automatically extract needed part from a metamodel. By deeply analyzing the characteristics such as special relations between packages and step-by-step strictly defining mechanism of modeling concepts, this paper presents an approach to pruning metamodels like UML as needed. The approach can effectively prune metamodels, control the size of pruned metamodels, and make pruned metamodels comply with its initial metamodels.
DNN (Deep Neural Networks) has achieved great success in various fields, but its deployment on many small devices is limited due to its huge model structure and low computing speed. The neural network faces the proble...
详细信息
ISBN:
(纸本)9783031067884;9783031067877
DNN (Deep Neural Networks) has achieved great success in various fields, but its deployment on many small devices is limited due to its huge model structure and low computing speed. The neural network faces the problems of large models and slow calculations. Experiments have shown that reasonable pruning methods can be effectively solved. In terms of network pruning, pruning technology can be divided into structured pruning and unstructured pruning. Compared with the limited application of unstructured pruning, structured pruning can greatly compress the model and increase the calculation speed under any framework, and has stronger applicability. In current structured pruning, there is a problem that the accuracy of the model decreases too quickly after a large number of neurons are deleted. To address this problem, we propose a new pruning method based on neuron similarity, which sorts the weights of neurons, according to the principle that neural network parameters will change with training, the integration method is introduced to reconstruct the neuron assignment, and the more relevant neurons are deleted by comparing the current ranking and the cumulative ranking difference of the integration system. Based on the experiments of MLP model, compared with other pruning methods, this method shows its superiority, which can compress the model by 10 times and reduce the accuracy by less than 1%.
In this paper, a new pruning method of digital predistortion based on optimal brain surgeon (OBS) algorithm is proposed, which reduces the algorithm complexity significantly in power scalable applications. The convent...
详细信息
ISBN:
(纸本)9781665435277
In this paper, a new pruning method of digital predistortion based on optimal brain surgeon (OBS) algorithm is proposed, which reduces the algorithm complexity significantly in power scalable applications. The conventional OBS algorithm is modified for complex value DPD application by pruning the redundant items according to their impact on model accuracy such as normalized mean squared error (NMSE). Compared to conventional DPD, more than 60% of the coefficients can be reduced almost without obvious performance degradation.
Echo state networks (ESNs) have become one of the most effective dynamic neural networks because of its excellent fitting performance in real-valued time series modeling tasks and simple training processes. The origin...
详细信息
Echo state networks (ESNs) have become one of the most effective dynamic neural networks because of its excellent fitting performance in real-valued time series modeling tasks and simple training processes. The original ESN concept used randomly fixed created reservoirs, and this concept is considered to be one of its main advantages. However, ESNs have been criticized for its randomly created connectivity and weight parameters. Determining the appropriate weight parameters for a given task is an important problem. An optimization method based on mutual information (MI) is proposed in this study to optimize the input scaling parameters and the structure of ESN to address the aforementioned problem and improve the performance of ESN. The MI optimization method mainly consists of two parts: First, the scaling parameters of multiple inputs are adjusted based on the MI between the network inputs and outputs. Second, some output weight connections are pruned for optimization based on the MI between reservoir states. Furthermore, three MI-ESN models are proposed for a fed-batch penicillin fermentation process. Our experimental outcomes reveal that the obtained MI-ESN models outperform the ESN models without optimization and other traditional neural networks.
For a directed graph G with vertex set V, we call a subset a k-(All-)Path Cover if C contains a node from any simple path in G consisting of k nodes. This paper considers the problem of constructing small k-Path Cover...
详细信息
For a directed graph G with vertex set V, we call a subset a k-(All-)Path Cover if C contains a node from any simple path in G consisting of k nodes. This paper considers the problem of constructing small k-Path Covers in the context of road networks with millions of nodes and edges. In many application scenarios, the set C and its induced overlay graph constitute a very compact synopsis of G, which is the basis for the currently fastest data structure for personalized shortest path queries, visually pleasing overlays of subsampled paths, and efficient reporting, retrieval and aggregation of associated data in spatial network databases. Apart from a theoretic investigation of the problem, we provide efficient algorithms that produce very small k-Path Covers for large real-world road networks (with a posteriori guarantees via instance-based lower bounds). We also apply our algorithms to other (social, collaboration, web, etc.) networks and can improve in several instances upon previous approaches.
Holiday passenger flow forecasting is essential to transportation plan-making and passenger flow organization in metro systems during holidays. Usually, daily passenger flow characteristics show a great difference bet...
详细信息
Holiday passenger flow forecasting is essential to transportation plan-making and passenger flow organization in metro systems during holidays. Usually, daily passenger flow characteristics show a great difference between holidays and normal days, and the annual growth of holiday passenger flow seems more complicated. Least-square support vector machine (LSSVM) is able to handle the complex fluctuations in holiday daily passenger flow, but it suffers from critical parameter selection, and sparseness is also lost in the LSSVM solution. In an attempt to forecast holiday passenger flow accurately, this paper proposes an approach based on the modified LSSVM, in which an improved particle-swarm optimization (IPSO) algorithm is developed to optimize parameters and pruning algorithm is used to achieve sparseness, as well as a new evaluation indicator based on the k-fold cross-validation method to evaluate the training performance. Finally, passenger flow data for Guangzhou Metro stations in China during the National Day holiday from 2011 to 2014 are applied as numerical examples to validate the performance of the proposed approach. The results show that the modified LSSVM model is an effective forecasting approach with higher accuracy than other alternative models. (C) 2016 American Society of Civil Engineers.
This paper presents an algorithm to extract symbolic rules from trained artificial neural networks (ANNs), called ERANN. In many applications, it is desirable to extract knowledge from ANNs for the users to gain a bet...
详细信息
This paper presents an algorithm to extract symbolic rules from trained artificial neural networks (ANNs), called ERANN. In many applications, it is desirable to extract knowledge from ANNs for the users to gain a better understanding of how the networks solve the problems. Although ANN usually achieves high classification accuracy, the obtained results sometimes may be incomprehensible, because the knowledge embedded within them is distributed over the activation functions and the connection weights. This problem can be solved by extracting rules from trained ANNs. To do so, a rule extraction algorithm has been proposed in this paper to extract symbolic rules from trained ANNs. A standard three-layer feedforward ANN with four-phase training is the basis of the proposed algorithm. Extensive experimental studies on a set of benchmark classification problems, including breast cancer, iris, diabetes, wine, season, golfplaying, and lenses classification, demonstrates the applicability of the proposed method. Extracted rules are comparable with other methods in terms of number of rules, average number of conditions for a rule, and the rules accuracy. The proposed method achieved accuracy values 96.28, 98.67, 76.56, 91.01, 100, 100, and 100 for the above problems, respectively. It has been seen that these results are one of the best results comparing with results obtained from related previous studies.
In future resource constrained systems such as Industry 4.0 and Cyber-Physical Systems, thousands of resource constrained embedded devices communicate in order to reach a common objective. Information, then, is proces...
详细信息
In future resource constrained systems such as Industry 4.0 and Cyber-Physical Systems, thousands of resource constrained embedded devices communicate in order to reach a common objective. Information, then, is processed and transformed several times since it is acquired at physical level until it is received by user applications. Moreover, usually, each component in an current engineered system has almost no information about the other deployed modules. In that way, any malicious component may severely affect systems through the modification of operation data, as high-level applications can check the validity of received information in a hard way. Therefore, future technological components should only communicate with other trustworthy modules. Typically, solutions to address this problem are based on heavy technologies (such as certificates) which do not fulfil the requirements of embedded devices making up future systems. Thus, in this paper, a light computational solution to calculate the graph representing the trustworthy routes through which an application may obtain information is provided. Systems are represented as directed graphs, and untrustworthy branches are identified based on probability offered by a biparametric stochastic process (which is estimated by means an agent-based solution, consisting of a static and a mobile agent). Untrustworthy paths are also pruned following a hierarchical and scalable process. The proposed solution is based on the execution of some "elemental requests" whose response is known. Using the NS3 simulator, virtualization technologies and TAP bridges, the proposed algorithm is validated in a simulation scenario representing a real deployment.
In the working process of Double-Fed Wind Turbines (DFWT), it is very important to monitor and predict the temperature of the high-speed output shaft of the gearbox timely and effectively. Support vector machine has m...
详细信息
In the working process of Double-Fed Wind Turbines (DFWT), it is very important to monitor and predict the temperature of the high-speed output shaft of the gearbox timely and effectively. Support vector machine has more advantages in the temperature prediction of wind turbines. Least squares support vector machine is suitable for online prediction due to reducing the computational complexity of support vector machine. In order to solve the sparsity of least squares support vector machine, an improved least squares support vector machine based on pruning algorithm is proposed in this paper to predict the temperature of the high-speed output shaft of gearbox using the practical data of Double-Fed Wind Turbines. At the same time, in order to improve the prediction accuracy and to solve the problem of few links between different feature parameters in common normalization method, the paper uses the method of joint normalization to preprocess the data. The principal component analysis is used to reduce the dimension of the data. Particle swarm optimization algorithm is used to optimize the parameters of the pruning least squares support vector machine. The proposed model that is established in this paper is a new model to forecast the temperature of the high-speed output shaft. The results show that its prediction accuracy is higher than that of other algorithms.
Automatic modulation recognition is a major project in the field of radio cognition;however, the generalization ability of conventional models cannot satisfy practical applications. In order to improve the generalizat...
详细信息
Automatic modulation recognition is a major project in the field of radio cognition;however, the generalization ability of conventional models cannot satisfy practical applications. In order to improve the generalization performance of the deep learning model and increase its recognition efficiency, we propose a novel model: ElsNet (elastic convolutional neural network). This network designs a channel optimization module, by inputting the average pooling information of the feature map and the intrinsic parameters of the batch normalization layer, to dynamically optimize the connection relations between network neurons and enhance the generalization ability of the model. ElsNet achieves an accuracy of about 94% at signal-to-noise ratios of 0-20 dB. Subsequent experiments have also demonstrated that, the ElsNet has a satisfying performance in transferred data sets and a peak accuracy of 82% through transfer learning, which to a certain extent alleviates the problem that the current signal modulation recognition can only be applied to signals with the same modulation parameters as the training dataset and has poor performance in recognizing real signals with different modulation parameters.
暂无评论