The certainty factor-based neural network in continuousclassification domains

The integration of certainty factors (CFs) into the neural computing framework has resulted in a special artificial neural network known as the CFNet. This paper presents the cont-CFNet, which is devoted to classification domains where instances are described by continuous attributes. A new mathematical analysis on learning behavior, specifically linear versus nonlinear learning, is provided that can serve to explain how the cont-CFNet discovers patterns and estimates output probabilities. Its advantages in performance and speed have been demonstrated in empirical studies.     

Distributed computing methodology for training neural networks in an image-guided diagnostic application

Distributed computing is a process through which a set of computers connected by a network is used collectively to solve a single problem. In this paper, we propose a distributed computing methodology for training neural networks for the detection of lesions in colonoscopy. Our approach is based on partitioning the training set across multiple processors using a parallel virtual machine. In this way, interconnected computers of varied architectures can be used for the distributed evaluation of the error function and gradient values, and, thus, training neural networks utilizing various learning methods. The proposed methodology has large granularity and low synchronization, and has been implemented and tested. Our results indicate that the parallel virtual machine implementation of the training algorithms developed leads to considerable speedup, especially when large network architectures and training sets are used.     

Geometric neural computing

This paper shows the analysis and design of feedforward neural networks using the coordinate-free system of Clifford or geometric algebra. It is shown that real-, complex-, and quaternion-valued neural networks are simply particular cases of the geometric algebra multidimensional neural networks and that some of them can also be generated using support multivector machines (SMVMs). Particularly, the generation of radial basis function for neurocomputing in geometric algebra is easier using the SMVM, which allows one to find automatically the optimal parameters. The use of support vector machines in the geometric algebra framework expands its sphere of applicability for multidimensional learning. Interesting examples of nonlinear problems show the effect of the use of an adequate Clifford geometric algebra which alleviate the training of neural networks and that of SMVMs.     

Application of efficient composite methods for computing with certainty periodic orbits in molecular systems

Recently, we have proposed a technique for the computation of periodic orbits in molecular systems, based on the characteristic bisection method [Vrahatis et al., Comput. Phys. Commun. 138 (2001) 53]. The main advantage of the characteristic bisection method is that it converges with certainty within a given starting rectangular region. In this paper we further improve this technique by applying, on a surface of section of a Poincaré map, an iterative scheme based on the composition of the characteristic bisection method with other more rapid root-finding methods such as Newton's or Broyden's methods. Thus, the composite schemes compute rapidly with certainty periodic orbits of molecular systems. By applying these methods to the LiNC/LiCN molecular system we obtain promising results. We have reproduced previous results using considerable less CPU time. Also, we have located and computed new asymmetric families of periodic orbits.     

Reservoir computing approaches to recurrent neural network training

Echo State Networks and Liquid State Machines introduced a new paradigm in artificial recurrent neural network (RNN) training, where an RNN (the reservoir) is generated randomly and only a readout is trained. The paradigm, becoming known as reservoir computing, greatly facilitated the practical application of RNNs and outperformed classical fully trained RNNs in many tasks. It has lately become a vivid research field with numerous extensions of the basic idea, including reservoir adaptation, thus broadening the initial paradigm to using different methods for training the reservoir and the readout. This review systematically surveys both current ways of generating/adapting the reservoirs and training different types of readouts. It offers a natural conceptual classification of the techniques, which transcends boundaries of the current “brand-names” of reservoir methods, and thus aims to help in unifying the field and providing the reader with a detailed “map” of it.     

Application of neural network computing to thermal non-destructive evaluation

A methodological study on the use of neural networks for defect characterisation by means of a thermal method is presented. Neural networks are used here as defect classifiers, based on the infrared emission of the target object after heating. In this kind of application, there is a high degree of uncertainty in defect class boundaries due to several factors, such as the noise in the measurement, the uneven heating of the target object and the anisotropies in its thermal conductivity. For this reason, the classical 1 of N coding scheme during training did not provide satisfactory results. Much better results have instead been obtained using a smoother activation function for the output units during training. The non-destructive evaluation of material using neural networks proved extremely satisfactory, especially when compared to the classical procedures of thermographic analysis.     

面向类脑计算的脉冲神经网络研究

随着深度学习在训练成本、泛化能力、可解释性以及可靠性等方面的不足日益突出,类脑计算已成为下一代人工智能的研究热点。脉冲神经网络能更好地模拟生物神经元的信息传递方式,且具有计算能力强、功耗低等特点,在模拟人脑学习、记忆、推理、判断和决策等复杂信息方面具有重要的潜力。本文对脉冲神经网络从以下几个方面进行总结:首先阐述脉冲神经网络的基本结构和工作原理;在结构优化方面,从脉冲神经网络的编码方式、脉冲神经元改进、拓扑结构、训练算法以及结合其他算法这5个方面进行总结;在训练算法方面,从基于反向传播方法、基于脉冲时序依赖可塑性规则方法、人工神经网络转脉冲神经网络和其他学习算法这4个方面进行总结;针对脉冲神经网络的不足与发展,从监督学习和无监督学习两方面剖析;最后,将脉冲神经网络应用到类脑计算和仿生任务中。本文对脉冲神经网络的基本原理、编码方式、网络结构和训练算法进行了系统归纳,对脉冲神经网络的研究发展具有一定的积极意义。     

网格计算的理论分析及其应用

针对网格计算这一热门话题，文章对它的理论进行了详细的分析和说明，包括主要目的、基本功能、计算协议和应用等，并对这一技术的发展进行了展望。     

On-line scheme for computing rotation factors

An integrated radix-2 on-line algorithm for computing rotation factors for matrix transformations is presented. The inputs are in sign-and-magnitude, floating-point representation and the outputs can be used in on-line signed-digit or in parallel form. The exponents are computed using conventional arithmetic while the significands are processed using on-line algorithms. The conventional result is obtained by using an on-the-fly conversion scheme. The rotation factors are computed in 10 + n clock cycles for n-bit significands. The clock period is kept small by the use of redundant adder schemes and low-precision estimates. The implementation and performance of the algorithm are discussed and compared with other approaches.     

The application of neural networks to vibrational diagnostics for multiple fault conditions

Vibration analysis has long been used for the detection and identification of machine fault conditions. The specific characteristics of the vibration spectrum that are associated with common fault conditions are quite well known, e.g. the BPOR spectral component reflecting bearing defects and the peak at the rotational frequency in the vibration spectrum indicating the degree of imbalance. The typical use of these features would be to determine when a machine should be taken out of operation in the presence of deteriorating fault conditions. Reliable diagnostics of deteriorating conditions may however be more problematic in the presence of simultaneous fault conditions. This paper demonstrates that the presence of a bearing defect makes it impossible to determine the degree of imbalance based on a single vibration feature, e.g. the peak at rotational frequency. In such a case, it is necessary to employ diagnostic techniques that are suited to the parallel processing of multiple features. Neural networks are the best known technique to approach such a problem. The paper demonstrates that a neural classifier using the X and Y components of both the peak at rotational frequency and the peak at BPOR frequency as input features can reliably diagnose the presence of bearing defect and can at the same time indicate the degree of imbalance. Several different neural techniques are evaluated for this purpose. It is first shown how Kohonen feature maps can be applied to do an exploratory analysis on the data, and to implement reliable classification of different bearing conditions. Different supervised neural classification techniques are then evaluated for their ability to reliably model the degree of imbalance, while also identifying the presence of defects.     

The application of neural networks to the papermaking industry

This paper describes the application of neural network techniques to the papermaking industry, particularly for the prediction of paper "curl". Paper curl is an important quality measure that can only be measured reliably off-line after manufacture, making it difficult to control. Here, we predict, before paper manufacture from characteristics of the current reel, whether the paper curl will be acceptable and the level of curl. For both issues the case of predicting the probability that paper will be "out-of-specification" and that of predicting the level of curl, we include confidence intervals indicating to the machine operator whether the predictions should be trusted. The results and the associated discussion describe a successful application of neural networks to a difficult, but important, real-world task taken from the papermaking industry. In addition the techniques described are widely applicable to industry where direct prediction of a quality measure and its acceptability are desirable.     

Integrating Expert Systems and neural computing for decision support

Expert (or knowledge-based) systems are used today either as stand-alone or in conjunction with other computer-based information systems (CBIS) in thousands of organizations worldwide to provide decision support for solving problems that traditional information systems were unable to solve. Neural computing is an emerging promising technology with few successful applications. By integrating the two technologies one can achieve some improvements in the implementation of each as well as increase the scope of application. We provide a comprehensive framework and examples of such an integration.     

A dental application of neural network computing: Classification of complex electrical impedance measurements to aid root canal treatment

Electrical impedance measurements provide an alternative diagnostic technique to the use of radiographs for aiding dental root canal treatment. Analysis of impedance data was based on Complex NonLinear Least Squares (CNLS) regression with electrical circuits as models. Different equivalent circuits were required to model the data at various depths within root canals. Therefore, it was not valid to compare directly the parameter values obtained for the same electrical components when different circuits were used. This problem was solved with a neural computing approach based on supervised training of the backpropagation algorithm to classify the data. Two strategies were investigated. The first produced a network output which indicated the electrode depth within the canal. The second approach employed the neural network as a preprocessor to establish which equivalent circuit was appropriate for the CNLS. Tests were also carried out to determine the minimum number of input nodes required by a neural network for this dental application.     

A distributed computing tool for generating neural simulation databases

After developing a model neuron or network, it is important to systematically explore its behavior across a wide range of parameter values or experimental conditions, or both. However, compiling a very large set of simulation runs is challenging because it typically requires both access to and expertise with high-performance computing facilities. To lower the barrier for large-scale model analysis, we have developed NeuronPM, a client/server application that creates a "screen-saver" cluster for running simulations in NEURON (Hines & Carnevale, 1997). NeuronPM provides a user-friendly way to use existing computing resources to catalog the performance of a neural simulation across a wide range of parameter values and experimental conditions. The NeuronPM client is a Windows-based screen saver, and the NeuronPM server can be hosted on any Apache/PHP/MySQL server. During idle time, the client retrieves model files and work assignments from the server, invokes NEURON to run the simulation, and returns results to the server. Administrative panels make it simple to upload model files, define the parameters and conditions to vary, and then monitor client status and work progress. NeuronPM is open-source freeware and is available for download at http://neuronpm.homeip.net . It is a useful entry-level tool for systematically analyzing complex neuron and network simulations.     

Multilayered neural architectures evolution for computing sequences of orthogonal polynomials

This article presents an evolutionary algorithm to autonomously construct full-connected multilayered feedforward neural architectures. This algorithm employs grammar-guided genetic programming with a context-free grammar that has been specifically designed to satisfy three important restrictions. First, the sentences that belong to the language produced by the grammar only encode all valid neural architectures. Second, full-connected feedforward neural architectures of any size can be generated. Third, smaller-sized neural architectures are favored to avoid overfitting. The proposed evolutionary neural architectures construction system is applied to compute the terms of the two sequences that define the three-term recurrence relation associated with a sequence of orthogonal polynomials. This application imposes an important constraint: training datasets are always very small. Therefore, an adequate sized neural architecture has to be evolved to achieve satisfactory results, which are presented in terms of accuracy and size of the evolved neural architectures, and convergence speed of the evolutionary process.     

The multisynapse neural network and its application to fuzzyclustering

In this paper, a new neural architecture, the multisynapse neural network, is developed for constrained optimization problems, whose objective functions may include high-order, logarithmic, and sinusoidal forms, etc., unlike the traditional Hopfield networks which can only handle quadratic form optimization. Meanwhile, based on the application of this new architecture, a fuzzy bidirectional associative clustering network (FBACN), which is composed of two layers of recurrent networks, is proposed for fuzzy-partition clustering according to the objective-functional method. It is well known that fuzzy c-means is a milestone algorithm in the area of fuzzy c-partition clustering. All of the following objective-functional-based fuzzy c-partition algorithms incorporate the formulas of fuzzy c-means as the prime mover in their algorithms. However, when an application of fuzzy c-partition has sophisticated constraints, the necessity of analytical solutions in a single iteration step becomes a fatal issue of the existing algorithms. The largest advantage of FBACN is that it does not need analytical solutions. For the problems on which some prior information is known, we bring a combination of part crisp and part fuzzy clustering in the third optimization problem.