SEARCH · Search PubMed

Results for “Networks”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,729 records · Page 96Linked to original sources

PubMedDate not supplied

An efficient learning algorithm for improving generalization performance of radial basis function neural networks.

This paper presents an efficient recursive learning algorithm for improving generalization performance of radial basis function (RBF) neural networks. The approach combines the rival penalized competitive learning (PRCL) [Xu, L., Kizyzak, A. & Oja, E. (1993). Rival penalized competitive learning for clustering analysis, RBF net and curve detection, IEEE Transactions on Neural Networks, 4, 636-649] and the regularized least squares (RLS) to provide an efficient and powerful procedure for constructing a minimal RBF network that generalizes very well. The RPCL selects the number of hidden units of network and adjusts centers, while the RLS constructs the parsimonious network and estimates the connection weights. In the RLS we derived a simple recursive algorithm, which needs no matrix calculation, and so largely reduces the computational cost. This combined algorithm significantly enhances the generalization performance and the real-time capability of the RBF networks. Simulation results of three different problems demonstrate much better generalization performance of the present algorithm over other existing similar algorithms.

Algorithms↗

PubMed2001

Pattern classification by a condensed neural network.

Neural networks have come to the fore as potent pattern classifiers. More amenable to parallel computation, they are much faster than the nearest neighbor classifier (NN), which, however, has distinctly outperformed them in several applications. The purpose of this study is to investigate a condensed neural network that combines the classification speed of neural networks and the low error rate of the nearest neighbor classifier. This condensed network is a fast, accurate classifier of simple architecture and function: it consists of a set of generalized perceptrons that draw maximal hyperspherical boundaries centered on patterns of memory units, each circumscribing reference patterns of a single category. The generalized perceptrons carry out classification, assisted by sporadic nearest neighbor matching to patterns of a small reference set. We compare the condensed network to a high performance neural network pattern classifier (Kohonen) and to NN in experiments on hand-printed character recognition.

Neural Networks, Computer↗

PubMed2001

A Modified General Regression Neural Network (MGRNN) with new, efficient training algorithms as a robust 'black box'-tool for data analysis.

A Modified General Regression Neural Network (MGRNN) is presented as an easy-to-use 'black box'-tool to feed in available data and obtain a reasonable regression surface. The MGRNN is based on the General Regression Neural Network by D. Specht [Specht, D. (1991). A General Regression Neural Network. IEEE Transactions on Neural Networks, 2(6), 568-576], therefore, the network's architecture and weights are determined. The kernel width of each training sample is trained by two supervised training algorithms. These fast and reliable algorithms require four user-definable parameters, but are robust against changes of the parameters. Its generalization ability was tested with different benchmarks: intertwined spirals, Mackey-Glass time series and PROBEN1. The MGRNN provides two additional features: (1) it is trainable with arbitrary data as long as a suitable metric exists. Particularly, it is unnecessary to force the data structure to vectors of equal length; (2) it is able to compute the gradient of the regression surface as long as the gradient of the metric is definable and defined. The MGRNN solves common practical problems of common feed-forward networks.

Algorithms↗

PubMed2003

Predicting the behaviour of G-RAM networks.

A Generalising Random Access Memory (G-RAM) neuron is distinguished from conventional neuron models by the fact that its tolerance to departures in similarity from its training pattern is variable. Details of this are given in this paper as it affects the behaviour a class of digital probabilistic neural networks which have been achieving attention in the neural networks literature for some years now. Such systems are also called n-tuple systems, weightless systems or p-RAM systems. After reviewing the literature on such networks, a novel simple combinatoric analysis of the most likely behaviour of recursive GRAM networks is described. The best network performance, measured by a key parameter called 'radius of retrievability' (first defined by Wong and Sherrington [J. Phys. A 22 (1989) 2233] as the error in the input that still allows evolution of the dynamic network to the correct attractor state), is obtained with a training set composed of random data patterns. Increasing the size of the training set reduces this radius of retrievability in a predictable manner. Changing the nature of the training set to non-random patterns also reduces the radius of retrievability to an extent that we show can be estimated from a measure of the diversity of the elements of the training set (we refer to this as the 'mean intra-set Hamming distance of the training set'). As mentioned earlier the feature of G-RAMs (indicated by the G) is that there exists a generalization parameter which determines how far a neuron input vector can stray from a training input for the neuron to respond in the trained way. It is shown that when this generalization parameter is reduced, then the radius of retrievability is also reduced but it is then stable in the face of an increase in size, or change in nature, of the training set. This is a novel prediction of the behaviour of systems and of the robustness of such behaviour in the face of varying the size and correlation properties of the training set.

Algorithms↗

PubMedDate not supplied

Evolving neural networks to identify bent-double galaxies in the FIRST survey.

The FIRST (Faint Images of the Radio Sky at Twenty-cm) survey is an ambitious project scheduled to cover 10,000 square degrees of the northern and southern galactic caps. Until recently, astronomers associated with FIRST identified radio-emitting galaxies with a bent-double morphology through a visual inspection of images. Besides being subjective, prone to error and tedious, this manual approach is becoming increasingly infeasible: upon completion, FIRST will include almost a million galaxies. This paper describes the application of six methods of evolving neural networks (NNs) with genetic algorithms (GAs) to the identification of bent-double galaxies. The objective is to demonstrate that GAs can successfully address some common problems in the application of NNs to classification problems, such as training the networks, choosing appropriate network topologies, and selecting relevant features. We measured the overall accuracy of the networks using the arithmetic and geometric means of the accuracies on bent and non-bent galaxies. Most of the combinations of GAs and NNs perform equally well on our data, but using GAs to select feature subsets produces the best results, reaching accuracies of 90% using the arithmetic mean and 87% with the geometric mean. The networks found by the GAs were more accurate than hand-designed networks and decision trees.

Astronomy↗

PubMed2003

Incremental training of first order recurrent neural networks to predict a context-sensitive language.

In recent years it has been shown that first order recurrent neural networks trained by gradient-descent can learn not only regular but also simple context-free and context-sensitive languages. However, the success rate was generally low and severe instability issues were encountered. The present study examines the hypothesis that a combination of evolutionary hill climbing with incremental learning and a well-balanced training set enables first order recurrent networks to reliably learn context-free and mildly context-sensitive languages. In particular, we trained the networks to predict symbols in string sequences of the context-sensitive language [a(n)b(n)c(n); n > or = 1. Comparative experiments with and without incremental learning indicated that incremental learning can accelerate and facilitate training. Furthermore, incrementally trained networks generally resulted in monotonic trajectories in hidden unit activation space, while the trajectories of non-incrementally trained networks were oscillating. The non-incrementally trained networks were more likely to generalise.

Language↗

PubMedDate not supplied

Passive dendritic integration heavily affects spiking dynamics of recurrent networks.

According to dendritic cable theory, proximal synapses give rise to inputs with short delay, high amplitude, and short duration. In contrast, inputs from distal synapses have long delays, low amplitude, and long duration. Nevertheless, large scale neural networks are seldom built with realistically layered synaptic architectures and corresponding electrotonic parameters. Here, we use a simple model to investigate the spike response dynamics of networks with different electrotonic structures. The networks consist of a layer of neurons receiving a sparse feedforward projection from a set of inputs, as well as sparse recurrent connections from within the layer. Firing patterns are set in the inputs, and recorded from the neuron (output) layer. The feedforward and recurrent synapses are independently set as proximal or distal, representing dendritic connections near or far from the soma, respectively. Analyses of firing dynamics indicate that recurrent distal synapses tend to concentrate network activity in fewer neurons, while proximal recurrent synapses result in a more homogeneous activity distribution. In addition, when the feedforward input is regular (spiking or bursting) and asynchronous, the output is regular if recurrent synapses are more distal than feedforward ones, and irregular in the opposite configuration. Finally, the amplitude of network fluctuations in response to asynchronous input is lower if feedforward and recurrent synapses are electrotonically distant from one another (in either configuration). In conclusion, electrotonic effects reflecting different dendritic positions of synaptic inputs significantly influence network dynamics.

Action Potentials↗

PubMed2000

A comparison between two neural network rule extraction techniques for the diagnosis of hepatobiliary disorders.

Neural networks have been widely used as tools for prediction in medicine. We expect to see even more applications of neural networks for medical diagnosis as recently developed neural network rule extraction algorithms make it possible for the decision process of a trained network to be expressed as classification rules. These rules are more comprehensible to a human user than the classification process of the networks which involves complex nonlinear mapping of the input data. This paper reports the results from two neural network rule extraction techniques, NeuroLinear and NeuroRule applied to the diagnosis of hepatobiliary disorders. The dataset consists of nine measurements collected from patients in a Japanese hospital and these measurements have continuous values. NeuroLinear generates piece-wise linear discriminant functions for this dataset. The continuous measurements have previously been discretized by domain experts. NeuroRule is applied to the discretized dataset to generate symbolic classification rules. We compare the rules generated by the two techniques and find that the rules generated by NeuroLinear from the original continuously valued dataset to be slightly more accurate and more concise than the rules generated by NeuroRule from the discretized dataset.

Algorithms↗

PubMedDate not supplied

Bayesian applications of belief networks and multilayer perceptrons for ovarian tumor classification with rejection.

Incorporating prior knowledge into black-box classifiers is still much of an open problem. We propose a hybrid Bayesian methodology that consists in encoding prior knowledge in the form of a (Bayesian) belief network and then using this knowledge to estimate an informative prior for a black-box model (e.g. a multilayer perceptron). Two technical approaches are proposed for the transformation of the belief network into an informative prior. The first one consists in generating samples according to the most probable parameterization of the Bayesian belief network and using them as virtual data together with the real data in the Bayesian learning of a multilayer perceptron. The second approach consists in transforming probability distributions over belief network parameters into distributions over multilayer perceptron parameters. The essential attribute of the hybrid methodology is that it combines prior knowledge and statistical data efficiently when prior knowledge is available and the sample is of small or medium size. Additionally, we describe how the Bayesian approach can provide uncertainty information about the predictions (e.g. for classification with rejection). We demonstrate these techniques on the medical task of predicting the malignancy of ovarian masses and summarize the practical advantages of the Bayesian approach. We compare the learning curves for the hybrid methodology with those of several belief networks and multilayer perceptrons. Furthermore, we report the performance of Bayesian belief networks when they are allowed to exclude hard cases based on various measures of prediction uncertainty.

Bayes Theorem↗

PubMed2003

Inductive bias strength in knowledge-based neural networks: application to magnetic resonance spectroscopy of breast tissues.

The integration of symbolic knowledge with artificial neural networks is becoming an increasingly popular paradigm for solving real-world applications. The paradigm provides means for using prior knowledge to determine the network architecture, to program a subset of weights to induce a learning bias which guide network training, and to extract knowledge from trained networks. The role of neural networks then becomes that of knowledge refinement. It thus provides a methodology for dealing with uncertainty in the prior knowledge. We address the open question of how to determine the strength of the inductive bias of programmed weights; we present a quantitative solution which takes the network architecture, the prior knowledge, and the training data into consideration. We apply our solution to the difficult problem of analyzing breast tissue from magnetic resonance spectroscopy (MRS); the available database is extremely limited and cannot be adequately explained by expert knowledge alone.

Adult↗

PubMed1997

Combining a neural network with case-based reasoning in a diagnostic system.

This paper presents a new approach for integrating case-based reasoning (CBR) with a neural network (NN) in diagnostic systems. When solving a new problem, the neural network is used to make hypotheses and to guide the CBR module in the search for a similar previous case that supports one of the hypotheses. The knowledge acquired by the network is interpreted and mapped into symbolic diagnosis descriptors, which are kept and used by the system to determine whether a final answer is credible, and to build explanations for the reasoning carried out. The NN-CBR model has been used in the development of a system for the diagnosis of congenital heart diseases (CHD). The system has been evaluated using two cardiological databases with a total of 214 CHD cases. Three other well-known databases have been used to evaluate the NN-CBR approach further. The hybrid system manages to solve problems that cannot be solved by the neural network with a good level of accuracy. Additionally, the hybrid system suggests some solutions for common CBR problems, such as indexing and retrieval, as well as for neural network problems, such as the interpretation of the knowledge stored in a neural network and the explanation of reasoning.

Artificial Intelligence↗

PubMed1992

Introduction of a neuronal network as a tool for diagnostic analysis and classification based on experimental pathologic data.

A neuronal network, as well as uni- and multivariate statistics and a discriminant analysis were applied to a morphometric database of 58 cases with thyroid neoplasms and normal thyroid tissue. The ability to classify cases correctly according to their diagnosis was compared between the neuronal network and discriminant analysis. For all pairwise comparisons, classification by neuronal network was as least as good as classification by discriminant analysis. For some comparisons, the neuronal network provided more correct diagnoses than discriminant analysis. On the contrary, in a comparison between tumors which are not significantly different according to multivariate statistics, the network reclassifies only half of the cases correctly, whereas discriminant analysis falsely suggests the possibility of classifying cases with either diagnosis. Our results confirm a higher sensitivity of the neuronal network to the diagnostic information contained in the present morphometric database, and we will therefore use this concept for analysis and diagnostic classification in further morphometric studies.

Diagnosis, Computer-Assisted↗

PubMed1997

Visualization of clinical data with neural networks, case study: polycystic ovary syndrome.

In medicine, the use of neural networks has concentrated mainly on classification problems. Clinicians are often interested in knowing what a patient's status is compared with other similar cases. Compared with biostatistics neural networks have one major drawback: the reliability of the classification is difficult to express. Therefore, clear visualization of the measurements can be more helpful than the calculated probability of a disease. The self-organizing map is the most widely used neural network for data visualization. Although, visualization can be attached to almost any feed-forward network as well. In this paper, we describe a topology-preserving feed-forward network and compare it with the self-organizing map. The two neural network models are used in a case study on the diagnosis of polycystic ovary syndrome, which is a common female endocrine disorder characterized by menstrual abnormalities, hirsutism and infertility.

Adolescent↗

PubMed2000

Improving diagnostic accuracy using a hierarchical neural network to model decision subtasks.

A number of quantitative models including linear discriminant analysis, logistic regression, k nearest neighbor, kernel density, recursive partitioning, and neural networks are being used in medical diagnostic support systems to assist human decision-makers in disease diagnosis. This research investigates the decision accuracy of neural network models for the differential diagnosis of six erythematous-squamous diseases. Conditions where a hierarchical neural network model can increase diagnostic accuracy by partitioning the decision domain into subtasks that are easier to learn are specifically addressed. Self-organizing maps (SOM) are used to portray the 34 feature variables in a two dimensional plot that maintains topological ordering. The SOM identifies five inconsistent cases that are likely sources of error for the quantitative decision models; the lower bound for the diagnostic decision error based on five errors is 0.0140. The traditional application of the quantitative models cited above results in diagnostic error levels substantially greater than this target level. A two-stage hierarchical neural network is designed by combining a multilayer perceptron first stage and a mixture-of-experts second stage. The second stage mixture-of-experts neural network learns a subtask of the diagnostic decision, the discrimination between seborrheic dermatitis and pityriasis rosea. The diagnostic accuracy of the two stage neural network approaches the target performance established from the SOM with an error rate of 0.0159.

Chronic Disease↗

PubMedDate not supplied

Using backpropagation networks for the estimation of aqueous activity coefficients of aromatic organic compounds.

This research examined the applicability of using a neural network approach to the estimation of aqueous activity coefficients of aromatic organic compounds from fragmented structural information. A set of 95 compounds was used to train the neural network, and the trained network was tested on a set of 31 compounds. A comparison was made between the results and those obtained using multiple linear regression analysis. With the proper selection of neural network parameters, the backpropagation network provided a more accurate prediction of the aqueous activity coefficients for testing data than did regression analysis. This research indicates that neural networks have the potential to become a useful analytical technique for quantitative prediction of structure-activity relationships.

Hydrocarbons↗

PubMedDate not supplied

Adaptive configuring of radial basis function network by hybrid particle swarm algorithm for QSAR studies of organic compounds.

The configuring of a radial basis function network (RBFN) consists of selecting the network parameters (centers and widths in RBF units and weights between the hidden and output layers) and network architecture. The issues of suboptimum and overfitting, however, often occur in RBFN configuring. This paper presented a hybrid particle swarm optimization (HPSO) algorithm to simultaneously search the optimal network structure and parameters involved in the RBFN (HPSORBFN) with an ellipsoidal Gaussian function as a basis function. The continuous version of PSO was used for parameter training, while the modified discrete PSO was employed to determine the appropriate network topology. The proposed HPSORBFN algorithm was applied to modeling the inhibitory activities of substituted bis[(acridine-4-carboxamide)propyl]methylamines to murine P388 leukemia cells and the bioactivities of COX-2 inhibitors. The results were compared with those obtained from RBFNs with the parameters optimized by continuous PSO and by conventionally RBFN training the algorithm for a fixed network topology, indicating that the HPSO was competent for RBFN configuring in that it converged quickly toward the optimal solution and avoided overfitting.

Algorithms↗

PubMed1996

Statistical approach to neural network model building for gentamicin peak predictions.

Feed forward neural networks are flexible, nonlinear modeling tools that are an extension of traditional statistical techniques. The hypothesis that feed forward neural network models can be built in a similar fashion as a statistical model was tested. Feed forward neural network models were built using forward and backward variable selection, and zero to five hidden nodes, and tanh and linear transfer functions were used. Gentamicin serum concentrations were predicted as a model drug for testing these methods. Peak observations from 392 patients were used to train, test, and validate the feed forward neural network. Inputs were demographic and drug dosing information. Model selection was performed using the Akaike information criteria (AIC), Bayesian information criteria (BIC), and a method of stopped training. The models with lowest root mean square (rms) error were those with all 10 inputs and five hidden nodes. Average rms error in the validation set was lowest for stopped training (1.46), then AIC (1.51), and finally BIC (1.56). Larger models tended to result in the best predictions. Overfitting can occur in models that are too large, either by using too many nodes in the hidden layer (rms = 1.49) or by using too many inputs with little information associated with them (rms = 1.70). We conclude that neural networks can be built using a large number of parameters that have good predictive performance. Care must be used during training to avoid overfitting the data. A stopped training method resulted in the network with the lowest rms error.

Adult↗

PubMed1999

Emergent oscillations in a realistic network: the role of inhibition and the effect of the spatiotemporal distribution of the input.

We have simulated a network of 10,000 two-compartment cells, spatially distributed on a two-dimensional sheet; 15% of the cells were inhibitory. The input to the network was spatially delimited. Global oscillations frequently were achieved with a simple set of connectivity rules. The inhibitory neurons paced the network, whereas the excitatory neurons amplified the input, permitting oscillations at low-input intensities. Inhibitory neurons were active over a greater area than excitatory ones, forming a ring of inhibition. The oscillation frequency was modulated to some extent by the input intensity, as has been shown experimentally in the striate cortex, but predominantly by the properties of the inhibitory neurons and their connections: the membrane and synaptic time constants and the distribution of delays. In networks that showed oscillations and in those that did not, widely distributed inputs could lead to the specific recruitment of the inhibitory neurons and to near zero activity of the excitatory cells. Hence the spatial distribution of excitatory inputs could provide a means of selectively exciting or inhibiting a target network. Finally, neither the presence of oscillations nor the global spike activity provided any reliable indication of the level of excitatory output from the network.

Computer Systems↗