卷 28, 编号 2 (2019)

In this paper, the passivity and passification issue are considered for a class of dynamic memristor neural networks (DMNNs) with delays. Different from the models with respect to memristive neural networks in the literature, the flux-controlled dynamic memristors are used in the neurons and finite concentrated delays are accounted for in the interconnections. With the construction of suitable Lyapunov-Krasovskii functional (LKF), a novel passivity criteria, which involves the interconnection matrix, the delayed interconnection matrix and nonlinear memristor, is addressed in the form of linear matrix inequalities (LMIs) in the flux-charge domain. In addition, for state feedback passification, two procedures for designing passification controllers are proposed in terms of LMIs to guarantee that the considered DMNNs are passive. Finally, two examples are provided to illustrate the validity of the theoretical results.

A diagonally connected two-dimensional lattice is the objective of the research. The model draws interest because each of its spins has 6 connects as in a 3D lattice. On the other hand, the planarity of the model allows us to use a strict polynomial algorithm to find its partition function and other characteristics. The Kasteleyn-Fisher algorithm is employed to carry out a computer simulation which enables us to see how the heat capacity behaves with the increasing lattice dimensionality. Given a finite lattice dimensionality, it is impossible to draw a definite conclusion, yet there is every reason to believe that the heat capacity diverges logarithmically at the critical point.

Face detection and recognition plays an important role in many occasions. This study explored the application of convolutional neural network in face detection and recognition. Firstly, convolutional neural network was briefly analyzed, and then a face detection model including three convolution layers, four pooling layers, introduction layers and three fully connected layers was designed. In face recognition, the self-learning convolutional neural network (CNN) model for global and local extended learning and Spatial Pyramid Pooling (SPP)-NET model were established. LFW data sets were used as model test samples. The results showed that the face detection model had an accuracy rate of 99%. In face recognition, the self-learning CNN model had an accuracy rate of 94.9% accuracy, and the SPP-Net model had an accuracy rate of 92.85%. It suggests that the face detection and recognition model based on convolutional neural network has good accuracy, and the face recognition efficiency of self-learning CNN model was better, which deserves further research and promotion.

In this paper a framework of the heterogeneous system for an adaptive classification of multivariate numerical data is proposed. Considered data characterize the dynamics of the simulated responses of auditory nerve fibers as probability that represents the average of the ensemble for acoustic signals. The possibility to classify sound stimuli by means of analysis of the responses of the auditory nerve for the fibers with high and low spontaneous rates is examined. The aim of the study was to develop and implement a method for an adaptive pattern recognition in limited a priori information about their number and structure. The proposed model architecture consists of several units that generalize basic stages of a perceptual process that in turn corresponds neuro-symbolic information processing approach to the machine perception problem. Proposed method combines the advantages of the self-organizing maps and the radial basic function networks. This combination leads to a hybrid learning approach, which allow to provide the automatic classification of unlabeled data. According to the obtained results, the proposed approach showed better accuracy for several complex benchmark tests as well as for pure tones recognition by means of simulated auditory nerve fibers responses compared to k-means and single-linkage unsupervised classification strategies.

Hybrid optical two-dimensional (2D) codes due to their large cardinality and improved system performance for constant weight have recently been designed and studied for optical CDMA system. In this manuscript, an attempt has been made to analyze the performance of four different 2D codes from a communication point of view; that describe their use to support a system under certain conditions. These hybrid 2D codes employ (n, w, λ_a, λ_c) optical orthogonal coding for one dimension that is for time spreading and synchronized prime sequences (SPS), synchronized quadratic congruence sequences (SQC), prime code sequences (PC), and synchronized prime procession (SPP) for other dimension that is for wavelength hopping respectively. The distinguishing property amongst the four codes lies in their values of maximum auto- and cross- correlation function. It has been observed that with the constant value of weight, the code with the largest code length parameter has shown the best performance while the code with the least code length gives the poorest performance. Investigations reveal that SPS/OOC code performs better in comparison to other codes with constant weight and for different number of wavelengths and SQC/OOC results in poor performance due to lesser code length.