Search

Chua's theory of local activity shows that local activity is the origin of complexity, and the complexity can only occur on or near the stable locally active domain, referred to as Edge of Chaos (EOC). Very recently, for the voltage-controlled locally active memristors, a new physical concept dubbed "Edge of Chaos Kernel" (EOCK), which consists of a series combination between a negative resistance and a negative inductance and exhibits the EOC phenomenon of being stable yet potentially unstable, was defined and applied to the Hodgkin-Huxley neural circuit model. When an EOCK is coupled to a passive environment, its stability is disrupted, resulting in the emergence of action potentials, chaos and various complex phenomena. This paper proposes the dual version of the EOCK called as R - C EOCK, which consists of the parallel combination between a negative resistance and a negative capacitance. We show that the actual NbO memristor manufactured by NaMLab essentially belongs to a current-controlled locally active memristor which contains a R - C EOCK and gives the signature of its EOCK and EOC. We construct a second-order neuron based on the NbO memristor when connected in parallel with a passive capacitor, and further prove that only memristors endowed with an EOCK can generate action potential. On this basis, we construct a minimum cellular neural network with only 7 components based on two NbO memristor neurons and a passive coupling resistor, in which neuromorphic behaviors of static and dynamic pattern formation may emerge if and only if the single neuron has an EOCK and is poised on the EOC. The analysis in this paper explains the dynamic mechanism of Smale's paradox, in which two mathematically dead neurons coupled by a passive environment may become alive, under the same or different input current excitation, which are more in line with the actual biological neural networks. • Brains need memristors blessed with an Edge of Chaos Kernel (EOCK), the crown jewel of emerging complexity. This paper extended the concept of EOCK of the voltage-controlled memristors to the generic current-controlled memristors through Chua's theories. • Based on the actual S-type NbO memristor, an intrinsic current-controlled memristor, we constructed a second-order NbO memristive neuron circuit by connecting a passive capacitance in parallel with the memristor, and further verified that it can generate action potentials such as damping spiking, periodic spiking, and self-sustained oscillation, if and only if the memristor is endowed with an EOCK. • Furthermore, this paper constructed the simplest fourth-order memristive cellular neural network (CNN) with only 7 components, containing two twinborn NbO memristive neurons with a passive resistor, from which we found that when both neurons are in the resting state and possess an EOCK, the appropriate coupling resistances can make the CNN produce a surprising oscillation phenomenon known as the inexplicable Smale's paradox, while the NbO neurons become "alive". These findings not only reveal the mechanism of the above Smale's paradox in a basic memristive CNN (which can be considered as reaction diffusion equations), but also explain the static and dynamic pattern formation of the memristive CNNs, which can provide a theoretical basis for the design and analysis of the CNNs. [ABSTRACT FROM AUTHOR]

The brain neurons' electrical activities represented by Electroencephalogram (EEG) signals are the most common data for diagnosing Epilepsy seizure, which is considered a chronic nervous disorder that cannot be controlled medically using surgical operation or medications with more than 40 % of Epilepsy seizure case. With the progress and development of artificial intelligence and deep learning techniques, it becomes possible to detect these seizures over the observation of the non-stationary-dynamic EEG signals, which contain important information about the mental state of patients. This paper provides a concerted deep machine learning model consisting of two simultaneous techniques detecting the activity of epileptic seizures using EEG signals. The time-frequency image of EEG waves and EEG raw waves are used as input components for the convolution neural network (CNN) and recurrent neural network (RNN) with long- and short-term memory (LSTM). Two processing signal methods have been used, Short-Time Fourier Transform (STFT) and Continuous Wavelet Transformation (CWT), have been used for generating spectrogram and scalogram images with sizes of 77 × 75 and 32 × 32, respectively. The experimental results showed a detection accuracy of 99.57 %, 99.57 % using CWT Scalograms, and 99.26 %, 97.12 % using STFT spectrograms as CNN input for the Bonn University dataset and the CHB-MIT dataset, respectively. Thus, the proposed models provide the ability to detect epileptic seizures with high success compared to previous studies. [ABSTRACT FROM AUTHOR]

This paper focuses on the quasi-invariant set (QIS), global attracting set (GAS) and global exponential attracting set (GEAS) of competitive neural networks (CNNs) with time-varying and infinite distributed delays. For these purposes, based on the characteristics of nonnegative matrix and M -matrix, a new bidirectional delay integral inequality and a novel integro-differential inequality are first established. From the founded integral inequality, the existence conditions of the QIS and the GAS of the discussed system are obtained. Besides, the existence conditions of the GEAS are also given by the proposed integro-differential inequality, which gets rid of the construction of complex Lyapunov functions and functionals. The frameworks of the QIS, GAS and GEAS are also given. A numerical example is analyzed to confirm the validity of the obtained results in the end. • The quasi-invariant, global attracting and global exponential attracting sets for competitive neural networks with mixed time-varying delays are studied. • Some new integral and integro-differential inequalities are established based on the characteristics of nonnegative matrix and M-matrix. • Some effective criteria for the quasi-invariant, global attracting and global exponential attracting sets are established. • The global exponential attracting set is straightway obtained by the integro-differential inequality, which gets rid of the construction of complex Lyapunov functions and functionals. [ABSTRACT FROM AUTHOR]

With the continuous development of technology and networks, real-life interactions are gradually being abstracted into social networks for study. Social circles are a fundamental structural feature that is prevalent on social networks. Thus, exploring social circle structure plays an important role in revealing the characteristics of complex social networks and provides guidance for understanding social behavior in real life. For example, it can aid in precision marketing, personalized recommendation, and knowledge dissemination within social circles. One of the important means of identifying social network circles lies on community detection algorithms. However, real-world social networks are often dynamic and can be studied and analyzed by building dynamic networks, while existing dynamic network community detection methods tend to ignore the global structure information and time-series information of nodes. To address this problem, this paper proposes a dynamic network community detection algorithm based on graph convolutional neural networks and contrastive learning, which fully captures the adjacent characteristics between nodes based on the correlation information and leverages the feature smoothing strategy to efficiently extract node representations of dynamic networks under unsupervised scenario. Specifically, the proposed algorithm first utilizes node correlation based aggregation strategy to compute the feature matrix for single time-step of the dynamic network. Then, mutual information maximization is implemented based on cross-entropy between learned local and global representations. To reduce the computational overhead in the optimization process, an additional LSTM module is further equipped for updating the parameters of graph convolutional networks in each time-step. Additionally, a contrastive learning based network smoothing strategy is designed to minimize the feature differences between neighboring nodes. Comparative experiments demonstrate that the proposed algorithm achieves excellent performance on both synthetic and real networks. • Solved the issue of inaccurate results in mining dynamic network community structures caused by an overreliance on network topology. • Successfully addressed the difficulty of representing global structural information in dynamic network representation learning. • Experimental results demonstrate that smoothing feature differences between adjacent time-slice networks in community mining is effective. [ABSTRACT FROM AUTHOR]

Dengue is a mosquito-borne viral disease widely spread in tropical and subtropical regions. Its adverse impact on the human health and global economies cannot be overstated. In order to implement more effective vector control measures, mechanisms that can more accurately forecast dengue cases are needed more urgently than before. In this paper, a novel hybrid architecture which has the advantages of both convolutional neural networks and recurrent neural networks is being proposed to forecast weekly dengue incidence. The forecasting performance of this architecture reveals that the deep hybrid architecture outperforms other frequently used deep learning models in dengue forecasting tasks. We have also evaluated the proposed models against state-of-the-art studies in the literature, demonstrating that our proposed hybrid models utilizing recurrent networks with convolutional layers can provide a significant boost in dengue forecasting. • Hybrid deep learning models are developed for dengue incidence forecasting. • The models give excellent forecasting accuracy up to 87.72% for weekly cases. • Up to 4 weeks in advance forecasting is provided. • The effects of different number of deep layers and used features are discussed. [ABSTRACT FROM AUTHOR]

The Speech Emotion Recognition (SER) is a complex task because of the feature selections that reflect the emotion from the human speech. The SER plays a vital role and is very challenging in Human-Computer Interaction (HCI). Traditional methods provide inconsistent feature extraction for emotion recognition. The primary motive of this paper is to improve the accuracy of the classification of eight emotions from the human voice. The proposed MFF-SAug research, Enhance the emotion prediction from the speech by Noise Removal, White Noise Injection, and Pitch Tuning. On pre-processed speech signals, the feature extraction techniques Mel Frequency Cepstral Coefficients (MFCC), Zero Crossing Rate (ZCR), and Root Mean Square (RMS) are applied and combined to achieve substantial performance used for emotion recognition. The augmentation applies to the raw speech for a contrastive loss that maximizes agreement between differently augmented samples in the latent space and reconstructs the loss of input representation for better accuracy prediction. A state-of-the-art Convolution Neural Network (CNN) is proposed for enhanced speech representation learning and voice emotion classification. Further, this MFF-SAug method is compared with the CNN + LSTM model. The experimental analysis was carried out using the RAVDESS, CREMA, SAVEE, and TESS datasets. Thus, the classifier achieved a robust representation for speech emotion recognition with an accuracy of 92.6 %, 89.9, 84.9 %, and 99.6 % for RAVDESS, CREMA, SAVEE, and TESS datasets, respectively. • Speech emotion recognition • Feature fusion – time and frequency domain feature extraction • Augmentation techniques • Convolution Neural Networks • Improved accuracy of emotion classification [ABSTRACT FROM AUTHOR]

In this paper, we exploit the adaptive stochastic resonance effect in the convolutional neural network with threshold activation functions for enabling the back-propagation gradient computation. During training, the injection of noise into the threshold activation function makes it to be differentiable at forward pass, thus exact gradients of the loss function with respect to network parameters including the injected noise level can smoothly propagate at backforward pass. This training method effectively uses the non-zero noise to improve the performance of the designed convolutional neural network. In the testing phase, the differentiable activation function is decoupled into a set of threshold functions driven by mutually independent noise components, which endows a hardware-friendly feature of the designed neural network. Experiments show that the designed low-precision convolutional neural network has a comparable performance with the full-precision one with ReLU activation functions on two benchmark data sets. These results also extend the potential application of adaptive stochastic resonance into the convolutional neural network for image classification. • Exploring adaptive stochastic resonance effect in threshold convolutional neural networks • Propagating exact gradients at backforward pass via the noise-smoothed neurons • Enhancing the testing accuracy of the convolutional neural networks by the noise injection [ABSTRACT FROM AUTHOR]

Inhalable particulate matter with a diameter of less than 2.5 μm spatio-temporal prediction technology is an important tool for environmental governance in urban traffic congestion areas. A new Ensemble Graph Attention Reinforcement Learning Recursive Network is proposed to create a multi-data-driven spatio-temporal prediction method with excellent application value. The modeling process includes three basic steps. In step I, the graph attention network is used to effectively aggregate the spatio-temporal correlation characteristics of the original air pollutant data. In step II, the features extracted from the graph attention network are transferred to the long short-term memory network and the temporal convolutional network and the prediction models are constructed respectively. In step III, the reinforcement learning algorithm effectively analyzes the adaptability of the two different models to the data sets and realizes ensemble based on continuous optimization of weights. By comparing the experimental results of the listed cases, the following points can be summarized: (a) the graph attention network can effectively aggregate the spatio-temporal correlation characteristics of the original data and optimize the performance of the predictor. (b) The reinforcement learning algorithm effectively realizes the integration of several neural networks and improves the comprehensive adaptability and generalization capabilities of the model. (c) The proposed model in this paper has great application potential and value in spatial and temporal prediction and has achieved better performance than the other 25 benchmark models. • A new multi-index driven spatio-temporal PM2.5 prediction model is proposed. • GAT is applied to study the spatio-temporal correlation between different sites. • The Sarsa algorithm is used to build the ensemble learning model with the strongest ability to adapt to different data. [ABSTRACT FROM AUTHOR]

An effective air pollution prediction is of great significance to prevent and control air pollution and protect the health of residents. In order to improve the prediction accuracy of PM 2.5 , an innovative PM 2.5 concentration prediction and early warning system based on optimal feature extraction and intelligent optimization is developed in this study. First, a feedback variational modal decomposition algorithm is designed to decompose the PM 2.5 concentration sequence and fuzzy entropy is used to reconstruct the patterns of similar complexity. Then, Copula entropy is used to select the influencing factors with a high impact on PM 2.5. Next, the reconstructed components and influencing factors are inputted to three individual prediction models, including long short-term memory neural network, gated recurrent unit neural network, and temporal convolutional network, for training and multi-step short-term prediction. The results of the individual prediction models are nonlinearly combined by Gaussian process regression which is optimized by the multi-objective grey wolf optimization algorithm. Finally, the prediction results of different reconstructed components are nonlinearly integrated to obtain the final PM 2.5 prediction results. In an empirical study of two Chinese cities, the combined prediction model proposed in this study outperformed the other six comparative models in terms of prediction accuracy and stability. The experimental results prove that the hybrid prediction model proposed in this paper can make an effective prediction and early warnings of air pollution. [ABSTRACT FROM AUTHOR]

To realize the secure transmission of double images, this paper proposes a double image encryption algorithm based on convolutional neural network (CNN) and dynamic adaptive diffusion. This scheme is different from the existing double image encryption technology. According to the characteristics of digital image, we design a dual-channel (digital channel / optical channel) encryption method, which not only ensures the security of double image, but also improves the encryption efficiency and reduces the possibility of being attacked. First, a chaotic map is used to control the initial values of the 5D conservative chaotic system to enhance the security of the key. Secondary, in order to effectively resist known-plaintext attack and chosen-plaintext attack, we employ a chaotic sequence as convolution kernel of convolution neural network to generate plaintext related chaotic pointer to control the scrambling operation of two images. On this basis, a novel image fusion method is designed, which splits and fuses two images into two different parts according to the amount of information contained. In addition, a dual-channel image encryption scheme, optical encryption channel and digital encryption channel, is designed for the two parts after fusion. The former has better parallelism and higher encryption efficiency, while the latter has higher computational complexity and better encryption reliability. Especially in the digital encryption channel, a new dynamic adaptive diffusion method is designed, which is more flexible and secure than the existing encryption algorithm. Finally, numerical simulation and experimental analysis verify the feasibility and effectiveness of the scheme. [ABSTRACT FROM AUTHOR]