Search

This paper provided an information-theoretic framework for the performance comparison of different types of HTM-Spatial Pooler (SP) algorithms for the first time. There are two primary goals. The first goal is to measure SP's performance as a standalone component, and the second goal is to compute the SP performance on the whole performance of the HTM algorithm. For this purpose, four different SPs were introduced by making changes in each of the four parts of the SP algorithm. The SP algorithm was considered a black box. Three information-theoretic measures, i.e., Renyi mutual information, Renyi divergence, and Henze-Penrose divergence, were proposed to determine the similarities and differences between the input and output of the SP algorithm. So the accuracy of each method was computed by these information-theoretic measures. This paper was able to quantify the similarities and differences between the two SDRs in the SP algorithm, unlike previous papers that do experimentally. Then, the results were compared with the Overlap Score measure that was previously introduced and found that the results of the new measures were compatible with the previous one. Four different datasets: MNIST, Fashion-MNIST, Hotgym, and NYC-Taxi, were used to perform experiments. The performance of each SP algorithm in the whole HTM system was examined and analyzed using the accuracy and error percentage measures. Finally, it concluded that the best SP algorithm's efficiency leads to the best HTM algorithm.

This paper investigates the problem of neural network (NN) prescribed performance tracking control for a class of switched stochastic nonlinear pure-feedback systems which contain unknown nonlinear functions, unmeasured state variables, and unknown hysteresis input. It provides an adaptive NN controller which is not restricted to a particular type of hysteresis input. A general mathematical model is introduced to describe two kinds of hysteresis nonlinearities and to utilize in the control design producer. Other focus of this paper is on the performance constraint problem to avoid performance degradation and system damage in practical control systems. Prescribed performance control (PPC) and backstepping technique are thus synthesized to develop an adaptive NN output feedback tracking control scheme under deterministic switching signal. Regarding this concern, to cope with the cause of non-differentiable difficulties and complex deductions in traditional PPC, a new asymmetry error transformation is employed. Moreover, radial basis function NNs (RBFNNs) are applied to approximate the unknown nonlinear functions and to construct a NN nonlinear observer to estimate the immeasurable state variables. Based on Lyapunov stability theory, it is demonstrated that the proposed controller can guarantee that all signals in the closed-loop system are semiglobally uniformly ultimately bounded in probability and the tracking error converges to a small neighborhood of the origin with the prescribed performance bounds. Finally, two simulation examples are provided to confirm the advantages of the presented control design approach.

The purpose of this paper is to investigate the problem of adaptive neural network (NN) output-feedback tracking control for input saturated switched stochastic nonlinear systems in pure-feedback form with unmeasured states. In order to facilitate the controller design process, a dead zone-based model of saturation is implemented and radial basis function NNs (RBFNNs) are employed to approximate unknown nonlinear functions and to construct an NN switched nonlinear observer to cope with difficulties raised by the presence of immeasurable state variables. Based on the adaptive backstepping technique and Lyapunov function method, an adaptive NN output feedback control scheme is developed. Furthermore, it is proved that the proposed controller can provide that, under arbitrary deterministic switching, all signals in the closed-loop system are semiglobally uniformly ultimately bounded in probability and the tracking error converges to a small neighborhood of the origin. Finally, simulation examples are presented to validate the effectiveness of the proposed adaptive NN control approach.

Training a support vector machine (SVM) on large datasets is a slow daunting process. Further, SVM becomes slow in the testing phase, due to its large number of support vectors (SVs). This paper proposes an effective geometric algorithm based on construction of Delaunay triangulation (DT) algorithm using Quickhull algorithm with a novel strategy to exactly identify and extract the boundary data points laid between the two classes of a dataset, and later uses these most informative data points as a reduced dataset to solve various SVM algorithms and proposes new DT-SVM algorithms Two synthetic datasets with the size of 1K incrementally up to 500K datasets are generated to extensively verify the effectiveness of the proposed DT-SVM algorithms over various data sizes and for further assessment, the most efficient version of proposed DT-SVM is applied on well-known benchmark datasets from UCI Machine Learning Repository. Two variant of sequential minimization optimization (SMO) decomposition methods, in addition to Least Square form of SVM are implemented to present the scalability of new DT-SVM algorithms in linear/nonlinear separable/non-separable large low dimensional datasets. Moreover, the most efficient version of the proposed algorithm is compared to RCH-SK as a known geometric approach in the SVM literature. The results demonstrate that while the proposed approach improves the scalability of DT-SVM in large low dimensional datasets, it leads SVM algorithms to maintain the accuracy in an acceptable range with considerably lower time in both training and testing phases with using a noticeably fewer number of SVs.

A variety of screening approaches have been proposed to diagnose epileptic seizures, using electroencephalography (EEG) and magnetic resonance imaging (MRI) modalities. Artificial intelligence encompasses a variety of areas, and one of its branches is deep learning (DL). Before the rise of DL, conventional machine learning algorithms involving feature extraction were performed. This limited their performance to the ability of those handcrafting the features. However, in DL, the extraction of features and classification are entirely automated. The advent of these techniques in many areas of medicine, such as in the diagnosis of epileptic seizures, has made significant advances. In this study, a comprehensive overview of works focused on automated epileptic seizure detection using DL techniques and neuroimaging modalities is presented. Various methods proposed to diagnose epileptic seizures automatically using EEG and MRI modalities are described. In addition, rehabilitation systems developed for epileptic seizures using DL have been analyzed, and a summary is provided. The rehabilitation tools include cloud computing techniques and hardware required for implementation of DL algorithms. The important challenges in accurate detection of automated epileptic seizures using DL with EEG and MRI modalities are discussed. The advantages and limitations in employing DL-based techniques for epileptic seizures diagnosis are presented. Finally, the most promising DL models proposed and possible future works on automated epileptic seizure detection are delineated.

In this research, we are investigating Convolutional Neural Networks (CNN) and Stacked Auto Encoders (SAE) to classify EEG Motor Imagery signals. Also, we use Cohen Class Distribution (CCD) to calculate time and frequency features derived from EEG signals to feed to our network. Using this combination of CNN and SAE decrease the data dimensions. the best accuracy percentage according to our method, in an average manner, is 82%. The proposed approach was applied to the dataset IVa from BCI Competition III, a multichannel 2-class motor-imagery dataset obtained from 5 healthy subjects

Digital watermarking is a remarkable issue in the field of information security to avoid the misuse of images in multimedia networks. Although access to unauthorized persons can be prevented through cryptography, it cannot be simultaneously used for copyright protection or content authentication with the preservation of image integrity. Hence, this paper presents an optimized multipurpose blind watermarking in the Shearlet domain with the help of smart algorithms, including MLP and NSGA-II. In this method, four copies of the robust copyright logo are embedded in the approximate coefficients of Shearlet by using an effective quantization technique. Furthermore, an embedded random sequence as a semi-fragile authentication mark is effectively extracted from details by the neural network. Due to performing an effective optimization algorithm for selecting optimum embedding thresholds and also distinguishing the texture of blocks, the imperceptibility and robustness have been preserved. The experimental results reveal the superiority of scheme with regard to the quality of watermarked images and robustness against hybrid attacks over other state-of-the-art schemes. The average PSNR and SSIM of the dual watermarked images are 38 dB and 0.95, respectively; Besides, it can effectively extract the copyright logo and locates forgery regions under severe attacks with satisfactory accuracy.

This paper presents an efficient hardware-based approach for segmentation of brain MR images utilizing kernel Fuzzy-C-Means (KFCM). First, the performance of the FCM algorithm for MR image segmentation is evaluated on 217* 181 images of BrainWeb dataset by a software implementation of it. Experiments demonstrated that the proposed algorithm provides a clustering accuracy of 96%; an appropriate result compared to other methods. After asserting the validity of the algorithm, the hardware implementation of the algorithm is conducted. Here, an overall architecture is proposed and designed toward being implemented on Xilinx FPGA Virtex7. Based on synthesis results, this design occupies approximately 19% of the hardware resources of xa7al00tcsg324-2I. Base obtained clock frequency of the designed system was 108MHz; then it is increased to 254 MHz by applying optimization techniques such as pipelining and reducing critical path delay. In the final proposed design, the amount of speed up factor compared to software implementation is 101.06. This improvement in speed is the aim of optimized hardware implementation.

In this paper, a new scheme for diagnosis of epileptic seizures in EEG signals using Tunable-Q wavelet transform (TQWT) framework is proposed and benchmarked with Bonn dataset. First, a segmentation of the EEG signals into smaller windows is performed, then a high-pass Butterworth filter with a cutoff frequency of 0.5 Hz is applied to eliminate possible noise. After that, a TQWT with parameters J=8, r=3, Q=1 has been utilized for decomposing the segmented EEG Signals into nine sub-bands. Next, a combination of statistical, entropy-based, and fractal dimension features are extracted from each sub-band. Finally, different ensemble learning-based classifiers, specifically, Adaboost, Gradient Boosting (GB), Hist Gradient Boosting (HistGB), and Random Forest (RF) are employed to classify signals. Also, a feature ranking is driven from classifiers to further analyze the importance of each feature in this particular task. Comparing our method to previous ones, introduced scheme outperforms most of the state-of-the-art works in this field, indicating the effectiveness of the proposed epileptic seizures detection method.

This paper presents a new approach for an online solution of the inverse kinematics problem based on nonlinear optimization for robots with joint physical constraints. The inverse kinematics problem is stated as a constrained nonlinear optimization problem and is solved using Kuhn--Tucker conditions analysis. The nonlinear multivariable optimization problem is locally converted into independent local linear constrained subproblems and each subproblem is analytically solved. For each joint, position limits and velocity limits are considered as physical constraints. The proposed structure benefits from a very low complexity design. The proposed method is fast and requires few calculations. The convergence of the proposed algorithm is proven based on the Lyapunov function. While keeping the algorithm stable, it can navigate the manipulator to a desired position under joint physical limits. The algorithm is simulated on a 7-DOF PA-10 manipulator. Results indicate good efficiency for the proposed structure in constrained path planning of joint space.

This paper addresses a new approach through detecting the P300 and its application to the BCI speller systems. This research employed stacked autoencoders which is based on many autoencoders and a classifier that is regularly a Softmax. This deep structure, decrease the dimension of the data and eventually, the reduced features of the last autoencoder are passed to the Softmax classifier. Subsequently, the parameters of the network would be ameliorated through a fine-tuning phase. Chebyshev Type I, is employed for filtering the EEG signals and using them as an input to the deep neural network. Hyperparameters such as the number of neurons and layers are attained empirically. Therefore, the final structure of the proposed network is 420-210-100-50-20-10-2. To analyze the suggested structure, the second dataset of the third BCI Competition is employed. According to the results, this approach can willingly enhance the character recognition in the BCI speller systems. Thus, the best accuracy percentage according to this research, in an average manner, is 91.5% of both A and B subjects. Consequently, according to the achievements, this method can be comparable to the other state-of-the-art algorithms and, therefore, can improve the recognition rate in the BCI industry.

Classification problems in a sequential framework is considered as an important field in pattern recognition. One of the main concerns in these types of problems is overtraining. In metacognitive neural networks (McNN), overtraining could be avoided by using the confidence of classifier (CoC) measure, which assigns a value between [0], [1] to class label. In this paper, a more accurate measure of CoC for McNN is presented. In addition, the hinge loss function which has no particular probabilistic interpretation is replaced by cross-entropy loss, which the output layer of the Soft meta-cognitive neural network (SMcNN) is a SoftMax layer. The classification performance is improved by applying the proposed SMcNN to well-known vCI datasets and comparing the results to McNN, SVM, and some other classifiers.

This paper presents new Radial Basis Function (RBF) learning methods for classification problems. The proposed methods use some heuristics to determine the spreads, the centers and the number of hidden neurons of network in such a way that the higher efficiency is achieved by fewer numbers of neurons, while the learning algorithm remains fast and simple. To retain network size limited, neurons are added to network recursively until termination condition is met. Each neuron covers some of train data. The termination condition is to cover all training data or to reach the maximum number of neurons. In each step, the center and spread of the new neuron are selected based on maximization of its coverage. Maximization of coverage of the neurons leads to a network with fewer neurons and indeed lower VC dimension and better generalization property. Using power exponential distribution function as the activation function of hidden neurons, and in the light of new learning approaches, it is proved that all data became linearly separable in the space of hidden layer outputs which implies that there exist linear output layer weights with zero training error. The proposed methods are applied to some well-known datasets and the simulation results, compared with SVM and some other leading RBF learning methods, show their satisfactory and comparable performance.

The support vector machine (SVM) is a powerful tool for classification problems. Unfortunately, the training phase of the SVM is highly sensitive to noises in the training set. Noises are inevitable in real-world applications. To overcome this problem, the SVM was extended to a fuzzy SVM by assigning an appropriate fuzzy membership to each data point. However, suitable choice of fuzzy memberships and an accurate model selection raise fundamental issues. In this paper, we propose a new method based on optimization methods to simultaneously generate appropriate fuzzy membership and solve the model selection problem for the SVM family in linear/nonlinear and separable/nonseparable classification problems. Both the SVM and least square SVM are included in the study. The fuzzy memberships are built based on dynamic class centers. The firefly algorithm (FA), a recently developed nature-inspired optimization algorithm, provides variation in the position of class centers by changing their attributes' values. Hence, adjusting the place of the class center can properly generate accurate fuzzy memberships to cope with both attribute and class noises. Furthermore, through the process of generating fuzzy memberships, the FA can choose the best parameters for the SVM family. A set of experiments is conducted on nine benchmarking data sets of the UCI data base. The experimental results show the effectiveness of the proposed method in comparison to the seven well-known methods of the SVM literature.

Several objective functions have been proposed in the literature to adjust the input parameters of a node in constructive networks. Furthermore, many researchers have focused on the universal approximation capability of the network based on the existing objective functions. In this brief, we use a correntropy measure based on the sigmoid kernel in the objective function to adjust the input parameters of a newly added node in a cascade network. The proposed network is shown to be capable of approximating any continuous nonlinear mapping with probability one in a compact input sample space. Thus, the convergence is guaranteed. The performance of our method was compared with that of eight different objective functions, as well as with an existing one hidden layer feedforward network on several real regression data sets with and without impulsive noise. The experimental results indicate the benefits of using a correntropy measure in reducing the root mean square error and increasing the robustness to noise.

Supp ort vector machine (SVM) has become one of the most popular machine-learning methods during the last years. The design of an efficient model and the proper adjustment of the SVMs parameters are integral to reducing the testing time and enhancing performance. In this paper, a new bipartite objective function consisted of the sparseness property and generalization performance is proposed. Since the proposed objective function is based on selecting fewer num- bers of the support vectors, the model complexity is reduced while the performance accuracy remains at an acceptable level. Due to the model complexity reduction, the testing time is decreased and the ability of SVM in practical applications is increased Moreover, to prove the performance of the proposed objective function, a comparative study was carried out on the proposed objective function and the conventional objective function, which is only based on the generalization perfor- mance, using the Binary Genetic Algorithm (BGA) and Real-valued vectors GA (RGA). The effectiveness of the proposed cost function is demonstrated based on the results of the comparative study on four real-world datasets of UCI database.

Power system security is one of the major concerns in recent years due to the deregulation of power systems which are forced to operate under stressed operating conditions. This paper presents an enhanced radial basis function neural network (ERBFNN) and winner-take-all neural network (WTANN) to examine whether the power system is secure under steady-state operating conditions. Hidden layer units have been selected with the proposed algorithm which has the advantage of being able to automatically choose optimal unit centers and distances. The proposed approach to contingency analysis was found to be suitable for fast voltage and line-flow contingency screening. The generalization capability of the proposed method was able to identify unknown contingencies with large range of operating conditions and changes in network topology. A feature extraction technique based on class separability index and correlation coefficient has been employed to identify the inputs and dimensional reduction for the ERBFNN and WTANN networks. The advantages of this method are simplicity of algorithm and high accuracy in classification. Case studies with IEEE 14-bus, IEEE 30-bus and IEEE 118-bus power systems are used to illustrate the good performance of the proposed method.

Timely and accurate assessment of voltage and power flow security is necessary to detect post-contingency problems in order to prevent a large-scale blackout. This article presents an enhanced radial basis function neural network based on a modified training algorithm for on-line ranking of the contingencies expected to cause steady-state bus voltage and power flow violations. Hidden layer neurons have been selected with the proposed algorithm, which has the advantage of being able to automatically choose optimal centers and radii. The proposed radial basis function neural network based security assessment algorithm has very small training time and space in comparison with multi-layer perceptron neural networks, support vector machines, and other machine learning based algorithms. A feature extraction technique based on kernel principal component analysis has been employed to identify the relevant inputs for the neural network. Also, the proposed feature extraction algorithm has been compared wit...