Data Generators for Learning Systems Based on RBF Networks

Geodesic Basis Function Neural Network

In the learning of existing radial basis function neural networks-based methods, it is difficult to propagate errors back. This leads to an inconsistency between the learning and recognition task. This article proposes a geodesic basis function neural network with subclass extension learning (GBFNN-ScE). The geodesic basis function (GBF), which is defined here for the first time, uses the geodetic distance in the manifold as a measure to obtain the response of the sample with respect to the local center. To learn network parameters by back-propagating errors for the purpose of correct classification, a specific GBF based on a pruned gamma encoding cosine function is constructed. This function has a concise and explicit expression on the hyperspherical manifold, which is conducive to the realization of error back propagation. In the preprocessing layer, a sample unitization method with no loss of information, nonnegative unit hyperspherical crown (NUHC) mapping, is proposed. The sample can be mapped to the support set of the GBF. To alleviate the problem that one-hot encoding is not effective enough in the differential expression of data labels within a class, a subclass extension (ScE) learning strategy is proposed. The ScE learning strategy can help the learned network be more robust. For the working of GBFNN-ScE, the original sample is projected onto the support set of GBF through the NUHC mapping. Then the mapped samples are sent to the nonlinear computation units composed of GBFs in the hidden layer. Finally, the response obtained in the hidden layer is weighted by the learned weight to obtain the network output. This article theoretically proves that the separability of the data with ScE learning is stronger. The experimental results show that the proposed GBFNN-ScE has a better performance in recognition tasks than existing methods. The ablation experiments show that the ideas of the GBFNN-ScE contribute to the algorithm performance.

Design of LDMOS Device Modeling Method Based on Neural Network

The rapid development of power semiconductor devices is helping to realize a low-carbon society and provide a better life for everyone. Power semiconductors not only are used in many large-scale industrial control fields such as power transmission and control in power grids, rail transit traction systems, and defense weapons and equipment, but also play a vital role in daily equipment such as home appliances, medical electronics, and electronic communications; all devices such as power steering in cars, battery chargers, cell phones, and microwave ovens utilize power electronics. This research mainly focuses on the high-voltage LDMOS device model and its implementation. Based on the in-depth study of the structure and physical mechanism of high-voltage LDMOS devices, with the help of BSIM4 core model, which is now very mature and widely used in industry, the drift region of high-voltage LDMOS is mainly modeled, and the drift region of LDMOS is modeled as a variable resistance controlled by voltage. Finally, Verilog-A language and neural network method are used to establish a compact model of LDMOS. The improved model is applied to LDMOS and can better fit the output characteristics with self-heating effect.

Convolutional Neural Network for Behavioral Modeling and Predistortion of Wideband Power Amplifiers

Power amplifier (PA) models, such as the neural network (NN) models and the multilayer NN models, have problems with high complexity. In this article, we first propose a novel behavior model for wideband PAs, using a real-valued time-delay convolutional NN (RVTDCNN). The input data of the model is sorted and arranged as a graph composed of the in-phase and quadrature ( I/Q ) components and envelope-dependent terms of current and past signals. Then, we created a predesigned filter using the convolutional layer to extract the basis functions required for the PA forward or reverse modeling. Finally, the generated rich basis functions are input into a simple, fully connected layer to build the model. Due to the weight sharing characteristics of the convolutional model's structure, the strong memory effect does not lead to a significant increase in the complexity of the model. Meanwhile, the extraction effect of the predesigned filter also reduces the training complexity of the model. The experimental results show that the performance of the RVTDCNN model is almost the same as the NN models and the multilayer NN models. Meanwhile, compared with the abovementioned models, the coefficient number and computational complexity of the RVTDCNN model are significantly reduced. This advantage is noticeable when the memory effects of the PA are increased by using wider signal bandwidths.

Bayesian Inference-Based Gaussian Mixture Models With Optimal Components Estimation Towards Large-Scale Synthetic Data Generation for In Silico Clinical Trials

Goal: To develop a computationally efficient and unbiased synthetic data generator for large-scale in silico clinical trials (CTs). Methods: We propose the BGMM-OCE, an extension of the conventional BGMM (Bayesian Gaussian Mixture Models) algorithm to provide unbiased estimations regarding the optimal number of Gaussian components and yield high-quality, large-scale synthetic data at reduced computational complexity. Spectral clustering with efficient eigenvalue decomposition is applied to estimate the hyperparameters of the generator. A case study is conducted to compare the performance of BGMM-OCE against four straightforward synthetic data generators for in silico CTs in hypertrophic cardiomyopathy (HCM). Results: The BGMM-OCE generated 30000 virtual patient profiles having the lowest coefficient-of-variation (0.046), inter- and intra-correlation differences (0.017, and 0.016, respectively) with the real ones in reduced execution time. Conclusions: BGMM-OCE overcomes the lack of population size in HCM which obscures the development of targeted therapies and robust risk stratification models.

A computational pipeline for data augmentation towards the improvement of disease classification and risk stratification models: A case study in two clinical domains

Virtual population generation is an emerging field in data science with numerous applications in healthcare towards the augmentation of clinical research databases with significant lack of population size. However, the impact of data augmentation on the development of AI (artificial intelligence) models to address clinical unmet needs has not yet been investigated. In this work, we assess whether the aggregation of real with virtual patient data can improve the performance of the existing risk stratification and disease classification models in two rare clinical domains, namely the primary Sjögren's Syndrome (pSS) and the hypertrophic cardiomyopathy (HCM), for the first time in the literature. To do so, multivariate approaches, such as, the multivariate normal distribution (MVND), and straightforward ones, such as, the Bayesian networks, the artificial neural networks (ANNs), and the tree ensembles are compared against their performance towards the generation of high-quality virtual data. Both boosting and bagging algorithms, such as, the Gradient boosting trees (XGBoost), the AdaBoost and the Random Forests (RFs) were trained on the augmented data to evaluate the performance improvement for lymphoma classification and HCM risk stratification. Our results revealed the favorable performance of the tree ensemble generators, in both domains, yielding virtual data with goodness-of-fit 0.021 and KL-divergence 0.029 in pSS and 0.029, 0.027 in HCM, respectively. The application of the XGBoost on the augmented data revealed an increase by 10.9% in accuracy, 10.7% in sensitivity, 11.5% in specificity for lymphoma classification and 16.1% in accuracy, 16.9% in sensitivity, 13.7% in specificity in HCM risk stratification.

A Sparse Neural Network Based Control Structure Optimization Game under DoS Attacks for DES Frequency Regulation of Power Grid

With the rapid growth of distributed energy sources, power grid has become a flexible and complex networked control system. However, it increases the chances of being a denial-of-service attack, which degrades the performance of the power grid, even causing cascading failures. To mitigate negative effects from denial-of-service attack and enhance the reliability of the power grid, we propose a networked control system structure based optimization scheme that is derived from a Stackelberg game model for the frequency regulation of a power grid with distributed energy sources. In the proposed game model, both denial-of-service attacker and control system designer as a defender are considered without using any analytical model. For defenders, we propose a sparse neural network based DES control and system structure design scheme. The neural network is used to approximate the desired control output and reinforce signals for the improvements of short- and long-term performance. It also introduces the sparse regulation of column grouping in the neural network learning process to explore the structure of control system that involves the placement of sensor, distributed energy sources actuator, and communication topology. For denial-of-service attackers, the related attack constraints and attack rewards are established. The solution of game equilibrium is considered as an optimal solution for both denial-of-service attack strategy and control structure. An offline optimization algorithm is proposed to solve the game equilibrium. The effectiveness of proposed scheme is verified by two cases, which illustrate the optimal solutions of both control structure and denial-of-service attack strategy.

Augmented Real-Valued Time-Delay Neural Network for Compensation of Distortions and Impairments in Wireless Transmitters

A digital predistorter, modeled by an augmented real-valued time-delay neural network (ARVTDNN), has been proposed and found suitable to mitigate the nonlinear distortions of the power amplifier (PA) along with modulator imperfections for a wideband direct-conversion transmitter. The input signal of the proposed ARVTDNN consists of Cartesian in-phase and quadrature phase ( I/Q ) components, as well as envelope-dependent terms. Theoretical analysis shows that the proposed model is able to produce a richer basis function containing both the desired odd- and even-order terms, resulting in improved modeling capability and distortion mitigation. Its actual performance has been validated through extensive simulations and experiments. The results show that the compensation and hardware impairment mitigation capabilities of the ARVTDNN are superior to the existing state-of-the-art real-valued focused time-delay neural network (RVFTDNN) by 3-4 dB for the adjacent channel power ratio and by 2-3 dB in terms of the normalized mean square error. Other important features of the proposed model are its reduced complexity, in terms of the number of parameters and floating-point operations, and its improved numerical stability compared to the RVFTDNN model.

Haze Removal Using Radial Basis Function Networks for Visibility Restoration Applications

Restoration of visibility in hazy images is the first relevant step of information analysis in many outdoor computer vision applications. To this aim, the restored image must feature clear visibility with sufficient brightness and visible edges, while avoiding the production of noticeable artifacts. In this paper, we propose a haze removal approach based on the radial basis function (RBF) through artificial neural networks dedicated to effectively removing haze formation while retaining not only the visible edges but also the brightness of restored images. Unlike traditional haze-removal methods that consist of single atmospheric veils, the multiatmospheric veil is generated and then dynamically learned by the neurons of the proposed RBF networks according to the scene complexity. Through this process, more visible edges are retained in the restored images. Subsequently, the activation function during the testing process is employed to represent the brightness of the restored image. We compare the proposed method with the other state-of-the-art haze-removal methods and report experimental results in terms of qualitative and quantitative evaluations for benchmark color images captured in typical hazy weather conditions. The experimental results demonstrate that the proposed method is able to produce brighter and more vivid haze-free images with more visible edges than can the other state-of-the-art methods.

Improving Neural-Network Classifiers Using Nearest Neighbor Partitioning

This paper presents a nearest neighbor partitioning method designed to improve the performance of a neural-network classifier. For neural-network classifiers, usually the number, positions, and labels of centroids are fixed in partition space before training. However, that approach limits the search for potential neural networks during optimization; the quality of a neural network classifier is based on how clear the decision boundaries are between classes. Although attempts have been made to generate floating centroids automatically, these methods still tend to generate sphere-like partitions and cannot produce flexible decision boundaries. We propose the use of nearest neighbor classification in conjunction with a neural-network classifier. Instead of being bound by sphere-like boundaries (such as the case with centroid-based methods), the flexibility of nearest neighbors increases the chance of finding potential neural networks that have arbitrarily shaped boundaries in partition space. Experimental results demonstrate that the proposed method exhibits superior performance on accuracy and average f-measure.