2019 Vol. 40, No. 11

Display Method:
CAO Jinde, SONG Qiankun, LIU Qingshan
2019, 40(11): 1-2.
Abstract(422) HTML (54) PDF(965)
An Exponential Laplace Loss Function Based Robust ELM for Regression Estimation
WANG Kuaini, CAO Jinde, LIU Qingshan
2019, 40(11): 1169-1178. doi: 10.21656/1000-0887.400240
Abstract(907) HTML (109) PDF(523)
Datasets are often contaminated by various noises in many practical applications. The classical extreme learning machine (ELM) shows poor prediction accuracy and large fluctuation of prediction results in dealing with such datasets. To overcome this drawback, an exponential Laplace loss function was proposed, which can weaken the influences of noises. The proposed loss function is based on the Gauss kernel function, and is differentiable, non-convex, bounded and able to approach the Laplace function. Then the proposed loss function was introduced into the ELM to build a robust ELM model for regression estimation. The iterative re-weighted algorithm was employed to solve the resultant optimization problem. In each iteration, the training samples with noises were given smaller weights, which can effectively improve the prediction accuracy. Experiments on real-world datasets show that, the proposed model has better learning performance and robustness.
Recent Progress of Approximate Bayesian Computation and Its Applications
ZHU Wanchuang, JI Chunlin, DENG Ke
2019, 40(11): 1179-1203. doi: 10.21656/1000-0887.400245
Abstract(1066) HTML (123) PDF(401)
In the era of big data and artificial intelligence, it is a common challenge for applied mathematics, statistics and computer science to extract valuable information and knowledge from complex data and models. Generative models are a class of powerful models which can potentially handle the above difficulty. From a macro point of view, the differential equations and dynamic systems in applied mathematics, the probability distribution in statistical models, and the typical generative models (generative adversarial networks and variational auto-encoders) in computer science could be considered as generalized generative models. Along with larger and larger-size data, the structure of data becomes more and more complicated simultaneously. Therefore, more powerful generative models are essential to process real problems. It is a challenge to describe mathematical structures of these generative models. It poses a natural question of how to analyze such generative models without analytic forms (or hard to obtain their analytic forms). Originated from the Bayesian inference, the approximate Bayesian computation, as a likelihood-free technique, plays an important role in processing complex statistical models and generative models. Based on the classic approximate Bayesian computation, the development and recent advance of approximate Bayesian computation were systematically reviewed. Finally, the application of the approximate Bayesian computation to complex data and the deep connection between the approximate Bayesian computation and cutting-edge artificial intelligence methods were discussed.
General Decay Synchronization for Recurrent Neural Networks With Distributed Time Delays
MUHAMMADHAJI Ahmadjan, LI Hongli
2019, 40(11): 1204-1213. doi: 10.21656/1000-0887.400127
Abstract(743) HTML (115) PDF(397)
The general decay synchronization (GDS) of a class of recurrent neural networks (RNNs) with general activation functions and distributed delays was studied. By means of suitable LyapunovKrasovskii functionals and useful inequality techniques, some sufficient conditions for the GDS of considered RNNs were established via a type of nonlinear control. An example with numerical simulations illustrates the correctness of the obtained theoretical results.
Adaptive Control of Nonlinear Systems Based on Quasi-ARX Multilayer Learning Network Models
WANG Lan, XIE Da, DONG Yiping, CAO Jinde
2019, 40(11): 1214-1223. doi: 10.21656/1000-0887.400212
Abstract(607) HTML (70) PDF(507)
A quasi-ARX multilayer learning network prediction model was established and applied to the adaptive control of nonlinear systems. The kernel of the model is an improved neuro-fuzzy network: one part is a 3-layer nonlinear network with an off-line training self-associative network, the other part is a 3-layer neuro-fuzzy network adjusted online. Accordingly, the parameters were classified and the corresponding estimation algorithms were given. Then, the controller design scheme was proposed based on the advantages of the macrostructure of the model. Simulation analysis verifies the effectiveness of the proposed model.
Global MittagLeffler Stability of Discrete-Time Fractional-Order Neural Networks
YOU Xingxing, LIANG Lunhai
2019, 40(11): 1224-1234. doi: 10.21656/1000-0887.400163
Abstract(1015) HTML (145) PDF(438)
The Mittag-Leffler stability of a class of discrete-time fractional-order neural networks was studied. Based on the discrete fractional calculus theory and the neural network theory, a class of discrete-time fractional-order neural networks were proposed. By means of the inequality techniques and the discrete Laplace transform, and through construction of the appropriate Lyapunov function, the sufficient criteria for global Mittag-Leffler stability of discrete-time fractional-order neural networks were obtained. Finally, a numerical simulation example verifies the validity of the proposed theory.
A Source Code Similarity Approach Based on Improved Convolutional Neural Networks
XIE Chunli, LIN Jiangxu, LIU Xiaoyang, ZHANG Wenbin, HUANG Junwei
2019, 40(11): 1235-1245. doi: 10.21656/1000-0887.400221
Abstract(844) HTML (107) PDF(378)
The source code similarity refers to the functional similarity of different code segments, which touches off important research in the field of software engineering. The existing methods mainly extracted texts and structure features manually from source codes to calculate the similarity based on the statistical information in disregard of the semantic characteristics of codes. To solve this problem, a source code similarity detection method based on the CNN was proposed. First, the source code was represented through word embedding to obtain the vector information of word embedding. Second, the CNN training model was constructed to learn the embedded representation of source code documents. Finally, the cosine similarity value of source code pairs was calculated. Experiments show that, the proposed method can certainly improve the performance with respect to the semantic similarity of source codes.
State Estimation of Complex-Valued Neural Networks With Leakage Delay and Mixed Additive Time-Varying Delays
LIU Libin, PAN Heping
2019, 40(11): 1246-1258. doi: 10.21656/1000-0887.400174
Abstract(705) HTML (101) PDF(330)
The state estimation of complex-valued neural networks with leakage delay and both discrete and distributed additive time-varying delays was studied. In the case where the activation function of the network was not required to be separated, through construction of the appropriate Lyapunov-Krasovskii functionals, and with the free weight matrix, the matrix inequality and the reciprocal convex combination method, the state of the neuron was estimated by means of observable output measurements. In addition, complex-valued linear matrix inequalities related to time delays were given to ensure the global asymptotic stability of the error-state model. Finally, numerical simulation examples verify the validity of the theoretical analysis.
A New Method for 3D TOA-Geolocation in Non-Line-of-Sight Environment
HAN Fengqing, XIAO Dan, GUAN Lihe
2019, 40(11): 1259-1269. doi: 10.21656/1000-0887.400191
Abstract(666) HTML (104) PDF(325)
The evolutionary game was introduced to reduce errors caused by the non-line-of-sight environment in 3D TOA-geolocation problems. A general dynamic replication model was established for the 3D TOA-geolocation problem in the non-line-of-sight environment. With each measuring base station as a player in this game, a TOA-geolocation algorithm based on the evolutionary game was proposed. The non-line-of-sight error was effectively reduced through iteration and the position of the mobile terminal was obtained finally. On the basis of the TOA-geolocation game algorithm, another 3D locating algorithm by means of the virtual base station and the virtual measuring data was designed for base stations with uneven position distribution. The experimental results show that, the TOA-geolocation algorithm based on the evolutionary game is slightly better than the classical algorithms. The 3D locating algorithm based on the virtual base station is more effective when the position distribution of the base station is obviously uneven.
Sparse Reconstruction of Fixed-Time Gradient Flow in the l1-l2 Norm
HU Dengzhou, HE Xing
2019, 40(11): 1270-1277. doi: 10.21656/1000-0887.400202
Abstract(805) HTML (110) PDF(319)
The compressed sensing (CS) is a new signal sampling technology, which can reconstruct signals at sampling points far smaller than those in the traditional Nyquist sampling theorem for sparse signals. For the compressed sensing, a dynamic continuous system was used to study the sparse signal reconstruction of the l1l-l2 norm. A sparse signal reconstruction algorithm based on the fixed time gradient flow was proposed, and was proved to be stable in the sense of Lyapunov and to converge to the optimal solution of the problem. Finally, the feasibility and advantages in the convergence speed of this algorithm were demonstrated through comparison between the proposed algorithm and existing projection neural network algorithms.
Optimal Leader-Follower Consensus of Multi-Agent Systems Based on the Event-Triggered Strategy
LIU Chen, LIU Lei
2019, 40(11): 1278-1288. doi: 10.21656/1000-0887.400216
Abstract(936) HTML (140) PDF(429)
The leader-follower consensus of linear multi-agent systems was investigated. An event-triggered adaptive dynamic programming method was proposed based on the undirected graph formed by means of the communication topology among agents, and the approximate optimal control was designed with the approximate properties of neural networks. According to the Lyapunov stability theorem, the stability of multi-agent error systems was analyzed, and a sufficient condition for the ultimate boundedness of the error system was found. Finally, numerical simulation results further verify the effectiveness of the theoretical analysis.
Expressway Travel Time Prediction Based on Gated Recurrent Unit Neural Networks
LIU Song, PENG Yong, SHAO Yiming, SONG Qiankun
2019, 40(11): 1289-1298. doi: 10.21656/1000-0887.400187
Abstract(685) HTML (92) PDF(479)
To efficiently predict the travel time on the expressway, the travel time was studied with the gated recurrent neural network through collection of the swiping data of vehicles at toll gates on the expressway. By means of the developed prediction computer program, the effects of the proposed method were then tested with the charging data of the Guangzhou Airport south expressway. The results show that the prediction effects are satisfying. Comparison with the LSTM neural network and the BP neural network indicates that, the gated recurrent neural network is better in prediction accuracy.
Finite-Time Combination Synchronization Control of Complex-Variable Chaotic Systems With Multi-Switching Transmission
LI Tianze, GUO Ming, CHEN Xiangyong, ZHANG Han, MA Jianyu
2019, 40(11): 1299-1308. doi: 10.21656/1000-0887.400206
Abstract(526) HTML (65) PDF(345)
The problem of finite-time combination synchronization for a class of complex-variable chaotic systems was investigated. Firstly, for the synchronization mode in signal transmission, the multi-switching synchronization behavior among multiple chaotic systems was analyzed. Secondly, based on the preset switching rules, the definition of finite-time combination synchronization was given. Then, according to the theory of finite-time stability, a kind of controller was designed to realize fast synchronization, and the sufficient conditions were given. Finally, results of numerical simulation and analysis verify the effectiveness of the proposed control scheme.