SEARCH · PubMed Health

Results for “Neural Networks, Computer”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

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

At least 1,693 records · Page 94Linked to original sources

PubMed2003

Neural computations in the tiger salamander and mudpuppy outer retinae and an analysis of GABA action from horizontal cells.

A neural network architecture based on the neural anatomy and function of retinal neurons in tiger salamander and mudpuppy retinae is proposed to study basic aspects of early visual information processing. The model predictions for the main response characteristics of retinal neurons are found to be in agreement with neurophysiological data, including the antagonistic role of horizontal cells in the outer plexiform layer. The examination of possible gamma-aminobutyric acid (GABA) action from horizontal cells suggests that GABA(A) alone, GABA(B) alone, or their weighted combination can generate the response characteristics observed in bipolar cells.

Ambystoma↗

PubMed2004

Inferring dynamic properties of biochemical reaction networks from structural knowledge.

Functional properties of biochemical networks depend on both the network structure and the kinetic parameters. Extensive data on metabolic network topologies have been collected in databases, but much less information is available about the kinetic constants or metabolite concentrations. Depending on the values of these parameters, metabolic fluxes and control coefficients may vary within a wide range. Nevertheless, some of the parameters may have little influence on the observables of interest. We address the question whether, despite uncertainty about kinetic parameters, probabilistic statements can be made about dynamic network features. To this end, we perform a variability analysis of the parameters: assuming that the parameters follow statistical distributions, we compute the resulting distributions of the network properties like metabolic fluxes, concentrations, or control coefficients by Monte Carlo simulation. In this manner, we study systematically the possible distributions arising from typical topologies of biochemical networks such as linear chains, branched networks, and signaling and gene expression cascades. This analysis reveals how much information about dynamic behavior can be drawn from structural knowledge.

Biochemistry↗

PubMed1994

Flexible vowel recognition by the generation of dynamic coherence in oscillator neural networks: speaker-independent vowel recognition.

We propose a new model for speaker-independent vowel recognition which uses the flexibility of the dynamic linking that results from the synchronization of oscillating neural units. The system consists of an input layer and three neural layers, which are referred to as the A-, B- and C-centers. The input signals are a time series of linear prediction (LPC) spectrum envelopes of auditory signals. At each time-window within the series, the A-center receives input signals and extracts local peaks of the spectrum envelope, i.e., formants, and encodes them into local groups of independent oscillations. Speaker-independent vowel characteristics are embedded as a connection matrix in the B-center according to statistical data of Japanese vowels. The associative interaction in the B-center and reciprocal interaction between the A- and B-centers selectively activate a vowel as a global synchronized pattern over two centers. The C-center evaluates the synchronized activities among the three formant regions to give the selective output of the category among the five Japanese vowels. Thus, a flexible ability of dynamical linking among features is achieved over the three centers. The capability in the present system was investigated for speaker-independent recognition of Japanese vowels. The system demonstrated a remarkable ability for the recognition of vowels very similar to that of human listeners, including misleading vowels. In addition, it showed stable recognition for unsteady input signals and robustness against background noise. The optimum condition of the frequency of oscillation is discussed in comparison with stimulus-dependent synchronizations observed in neurophysiological experiments of the cortex.

Auditory Cortex↗

PubMed2006

Event-driven simulation of neural population synchronization facilitated by electrical coupling.

Most neural communication and processing tasks are driven by spikes. This has enabled the application of the event-driven simulation schemes. However the simulation of spiking neural networks based on complex models that cannot be simplified to analytical expressions (requiring numerical calculation) is very time consuming. Here we describe briefly an event-driven simulation scheme that uses pre-calculated table-based neuron characterizations to avoid numerical calculations during a network simulation, allowing the simulation of large-scale neural systems. More concretely we explain how electrical coupling can be simulated efficiently within this computation scheme, reproducing synchronization processes observed in detailed simulations of neural populations.

Action Potentials↗

PubMed2004

Modeling segmentation of a visual scene via neural oscillators: fragmentation, discovery of details and attention.

The present study analyses the problem of binding and segmentation of a visual scene by means of a network of neural oscillators, laying emphasis on the problems of fragmentation, perception of details at different scales and spatial attention. The work is based on a two-layer model: a second layer of Wilson-Cowan oscillators is inhibited by information from the first layer. Moreover, the model uses a global inhibitor (GI) to segment objects. Spatial attention consists of an excitatory input, surrounded by an inhibitory annulus. A single object is identified by synchronous oscillatory activity of neural groups. The main idea of this work is that segmentation of objects at different detail levels can be achieved by linking parameters of the GI (i.e. the threshold and the inhibition strength) with the dimension of the zone selected by attention and with the dimension of the smaller objects to be detected. Simulations show that three possible kinds of behavior can be attained with the model, through proper choice of the GI parameters and attention input: (i) large objects in the visual scene are perceived, while small details are suppressed; (ii) large objects are perceived, while details are assembled together to constitute a single 'noise term'; (iii) if attention is focused on a smaller area and the GI parameters modulated accordingly (i.e. the threshold and attention strength are reduced) details are individually perceived as separate objects. These results suggest that the GI and attention may represent two concurrent aspects of the same attentive mechanism, i.e. they should work together to provide flexible management of a visual scene at different levels of detail.

Attention↗

PubMed1998

Reveal, a general reverse engineering algorithm for inference of genetic network architectures.

Given the immanent gene expression mapping covering whole genomes during development, health and disease, we seek computational methods to maximize functional inference from such large data sets. Is it possible, in principle, to completely infer a complex regulatory network architecture from input/output patterns of its variables? We investigated this possibility using binary models of genetic networks. Trajectories, or state transition tables of Boolean nets, resemble time series of gene expression. By systematically analyzing the mutual information between input states and output states, one is able to infer the sets of input elements controlling each element or gene in the network. This process is unequivocal and exact for complete state transition tables. We implemented this REVerse Engineering ALgorithm (REVEAL) in a C program, and found the problem to be tractable within the conditions tested so far. For n = 50 (elements) and k = 3 (inputs per element), the analysis of incomplete state transition tables (100 state transition pairs out of a possible 10(15)) reliably produced the original rule and wiring sets. While this study is limited to synchronous Boolean networks, the algorithm is generalizable to include multi-state models, essentially allowing direct application to realistic biological data sets. The ability to adequately solve the inverse problem may enable in-depth analysis of complex dynamic systems in biology and other fields.

Algorithms↗

PubMed2005

Training a learning vector quantization network using the pattern electroretinography signals.

In this study, the pattern electroretinography (PERG) signals derived from evoked potential across retinal cells of subjects after visual stimulation were analyzed using artificial neural network (ANN) with 172 healthy and 148 diseased subjects. ANN was employed to PERG signals to distinguish between healthy eye and diseased eye. Supervised network examined was a competitive learning vector quantization network. The designed classification structure has about 94% sensitivity, 90.32% specifity, 5.94% false negative, 9.67% false positive and correct classification is calculated to be 92%. Testing results were found to be compliant with the expected results that are derived from the physician's direct diagnosis. The end benefit would be to assist the physician to make the final decision without hesitation.

Adult↗

PubMed2004

Recurrent network with large representational capacity.

A recurrent network is proposed with the ability to bind image features into a unified surface representation within a single layer and without capacity limitations or border effects. A group of cells belonging to the same object or surface is labeled with the same activity amplitude, while cells in different groups are kept segregated due to lateral inhibition. Labeling is achieved by activity spreading through local excitatory connections. In order to prevent uncontrolled spreading, a separate network computes the intensity difference between neighboring locations and signals the presence of the surface boundary, which constrains local excitation. The quality of surface representation is not compromised due to the self-excitation. The model is also applied on gray-level images. In order to remove small, noisy regions, a feedforward network is proposed that computes the size of surfaces. Size estimation is based on the difference of dendritic inhibition in lateral excitatory and inhibitory pathways, which allows the network to selectively integrate signals only from cells with the same activity amplitude. When the output of the size estimation network is combined with the recurrent network, good segmentation results are obtained. Both networks are based on biophysically realistic mechanisms such as dendritic inhibition and multiplicative integration among different dendritic branches.

Action Potentials↗

PubMed2006

Fast numerical methods for simulating large-scale integrate-and-fire neuronal networks.

We discuss numerical methods for simulating large-scale, integrate-and-fire (I&F) neuronal networks. Important elements in our numerical methods are (i) a neurophysiologically inspired integrating factor which casts the solution as a numerically tractable integral equation, and allows us to obtain stable and accurate individual neuronal trajectories (i.e., voltage and conductance time-courses) even when the I&F neuronal equations are stiff, such as in strongly fluctuating, high-conductance states; (ii) an iterated process of spike-spike corrections within groups of strongly coupled neurons to account for spike-spike interactions within a single large numerical time-step; and (iii) a clustering procedure of firing events in the network to take advantage of localized architectures, such as spatial scales of strong local interactions, which are often present in large-scale computational models-for example, those of the primary visual cortex. (We note that the spike-spike corrections in our methods are more involved than the correction of single neuron spike-time via a polynomial interpolation as in the modified Runge-Kutta methods commonly used in simulations of I&F neuronal networks.) Our methods can evolve networks with relatively strong local interactions in an asymptotically optimal way such that each neuron fires approximately once in [Formula: see text] operations, where N is the number of neurons in the system. We note that quantifications used in computational modeling are often statistical, since measurements in a real experiment to characterize physiological systems are typically statistical, such as firing rate, interspike interval distributions, and spike-triggered voltage distributions. We emphasize that it takes much less computational effort to resolve statistical properties of certain I&F neuronal networks than to fully resolve trajectories of each and every neuron within the system. For networks operating in realistic dynamical regimes, such as strongly fluctuating, high-conductance states, our methods are designed to achieve statistical accuracy when very large time-steps are used. Moreover, our methods can also achieve trajectory-wise accuracy when small time-steps are used.

Algorithms↗

PubMed2001

Recruitment learning of boolean functions in sparse random networks.

This work presents a new class of neural network models constrained by biological levels of sparsity and weight-precision, and employing only local weight updates. Concept learning is accomplished through the rapid recruitment of existing network knowledge - complex knowledge being realised as a combination of existing basis concepts. Prior network knowledge is here obtained through the random generation of feedforward networks, with the resulting concept library tailored through distributional bias to suit a particular target class. Learning is exclusively local - through supervised Hebbian and Winnow updates - avoiding the necessity for backpropagation of error and allowing remarkably rapid learning. The approach is demonstrated upon concepts of varying difficulty, culminating in the well-known Monks and LED benchmark problems.

Algorithms↗

PubMed1995

Breast cancer: prediction with artificial neural network based on BI-RADS standardized lexicon.

PURPOSE: To determine if an artificial neural network (ANN) to categorize benign and malignant breast lesions can be standardized for use by all radiologists. MATERIALS AND METHODS: An ANN was constructed based on the standardized lexicon of the Breast Imaging Recording and Data System (BI-RADS) of the American College of Radiology. Eighteen inputs to the network included 10 BI-RADS lesion descriptors and eight input values from the patient's medical history. The network was trained and tested on 206 cases (133 benign, 73 malignant cases). Receiver operating characteristic curves for the network and radiologists were compared. RESULTS: At a specified output threshold, the ANN would have improved the positive predictive value (PPV) of biopsy from 35% to 61% with a relative sensitivity of 100%. At a fixed sensitivity of 95%, the specificity of the ANN (62%) was significantly greater than the specificity of radiologists (30%) (P < .01). CONCLUSION: The BI-RADS lexicon provides a standardized language between mammographers and an ANN that can improve the PPV of breast biopsy.

Adult↗

PubMed2004

Investigation of color constancy with a neural network.

The perceptual stability of an object's color under different illuminants is called color constancy. We created a neural network to investigate this phenomenon. The net consisted of one input channel for the background and one for the test object. Each channel had a set of three (L, M, and S) receptors that were transmitting to three opponent neurons. The signals from the opponent neurons were passed to hidden neurons, which were connected to the output neurons. The output signal was generated from the three components of a color vector. The neural net was trained to identify the color of Munsell samples under various illuminants using the back-propagation algorithm. Our study investigated the properties of a successfully trained neural network. Based on the cross-neuron weight analysis, we report that the successfully trained neural net calculates color differences between the test object and the background. By comparing the human visual system to the neural net, we conclude that to satisfy the color constancy phenomenon, the human visual system has to contain two separate components: one to approximate the background color and the other to estimate the color difference between the object and the background.

Color↗

PubMed2004

Biochemical Network Stochastic Simulator (BioNetS): software for stochastic modeling of biochemical networks.

BACKGROUND: Intrinsic fluctuations due to the stochastic nature of biochemical reactions can have large effects on the response of biochemical networks. This is particularly true for pathways that involve transcriptional regulation, where generally there are two copies of each gene and the number of messenger RNA (mRNA) molecules can be small. Therefore, there is a need for computational tools for developing and investigating stochastic models of biochemical networks. RESULTS: We have developed the software package Biochemical Network Stochastic Simulator (BioNetS) for efficiently and accurately simulating stochastic models of biochemical networks. BioNetS has a graphical user interface that allows models to be entered in a straightforward manner, and allows the user to specify the type of random variable (discrete or continuous) for each chemical species in the network. The discrete variables are simulated using an efficient implementation of the Gillespie algorithm. For the continuous random variables, BioNetS constructs and numerically solves the appropriate chemical Langevin equations. The software package has been developed to scale efficiently with network size, thereby allowing large systems to be studied. BioNetS runs as a BioSpice agent and can be downloaded from http://www.biospice.org. BioNetS also can be run as a stand alone package. All the required files are accessible from http://x.amath.unc.edu/BioNetS. CONCLUSIONS: We have developed BioNetS to be a reliable tool for studying the stochastic dynamics of large biochemical networks. Important features of BioNetS are its ability to handle hybrid models that consist of both continuous and discrete random variables and its ability to model cell growth and division. We have verified the accuracy and efficiency of the numerical methods by considering several test systems.

Algorithms↗

PubMed2001

Performance gain in computer-assisted detection schemes by averaging scores generated from artificial neural networks with adaptive filtering.

The authors investigated a new method to optimize artificial neural networks (ANNs) with adaptive filtering used in computer-assisted detection schemes in digitized mammograms and to assess performance changes when averaging classification scores from three sets of optimized schemes. Two independent training and testing image databases involving 978 and 830 digitized mammograms, respectively, were used in this study. In the training data set, initial filtering and subtraction resulted in the identification of 592 mass regions and 3790 suspicious, but actually negative regions. These regions (including both true-positive and negative regions) were segmented into three subsets three times based on the calculation of the values of three features as segmentation indices. The indices were "mass" size multiplied by their digital value contrast, conspicuity, and circularity. Nine ANN-based classifiers were separately optimized using a genetic algorithm for each subset of regions. Each region was assigned three classification scores after applying the three adaptive ANNs. The performance gain of the CAD scheme after averaging the three scores for each suspicious region was tested using an independent data set and a ROC methodology. The experimental results showed that the areas under ROC curves (Az) for the testing database using three sets of optimized ANNs individually were 0.84+/-0.01, 0.83+/-0.01, and 0.84+/-0.01, respectively. The between-index correlations of three A values were 0.013, -0.007, and 0.086. Similar to averaging diagnostic ratings from independent observers, by averaging three ANN-generated scores for each testing region, the performance of the CAD scheme was significantly improved (p<0.001) with Az value of 0.95+/-0.01.

Algorithms↗

PubMed1996

Propagation of spindle waves in a thalamic slice model.

1. We study the propagation and dynamics of spindle waves in thalamic slices by developing and analyzing a model of reciprocally coupled populations of excitatory thalamocortical (TC) neurons and inhibitory thalamic reticular (RE) neurons. 2. Each TC neuron has three intrinsic ionic currents: a low-threshold T-type Ca+2 current (ICa-T), a hyperpolarization-activated cation ("sag") current (Ih) and a leak current. Each RE cell also has three currents: ICa-T, a leak current, and a calcium-activated potassium current (IAHP). Isolated TC cells are at rest, can burst when released or depolarized from a hyperpolarized level, and burst rhythmically under moderate constant hyperpolarizing current. Isolated RE cells are at a hyperpolarized resting membrane potential and can burst when depolarized. 3. TC cells excite RE cells with fast alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) synapses, and RE cells inhibit TC cells with fast gamma-aminobutyric acid-A (GABAA) and slow GABAB synapses and inhibit each other with GABAA synapses only. GABAB postsynaptic conductances operate far from saturation, and the slow inhibitory postsynaptic potentials (IPSPs) increase with the width of the presynaptic burst. The model network is a one-dimensional cellular array with localized coupling. The synaptic coupling strength decays with the distance between the pre- and postsynaptic cells, either exponentially or as a step function. 4. The "intact" network can oscillate with partial synchrony and a population frequency of approximately 10 Hz. RE cells emit bursts almost at every oscillation cycle, whereas TC cells do so almost at every other cycle. Block of GABAB receptors hardly changes the network behavior. Block of GABAA receptors leads the network to a slowed oscillatory state, where the population frequency is approximately 4 Hz and both RE and TC cells fire unusually long bursts at every cycle and in full synchrony. These results are consistent with the experimental observations of von Krosigk, Bal, and McCormick. We obtain such consistency only when the above assumptions regarding the synaptic dynamics, particularly nonsaturating GABAB synapses, are fulfilled. 5. The slice model has a stable rest state with no neural activity. By initially depolarizing a few neurons at one end of the slice while all the other cells are at rest, a recruitment process may be initiated, and a wavefront of oscillatory activity propagates across the slice. Ahead of the wavefront, neurons are quiescent; neurons behind it oscillate. We find that the wave progresses forward in a lurching manner. TC cells that have just become inhibited must be hyperpolarized for a long enough time before they can fire rebound bursts and recruit RE cells. This step limits the wavefront velocity and may involve a substantial part of the cycle when no cells at the front are depolarized. 6. The wavefront velocity increases linearly with the characteristic spatial length of the connectivity (the footprint length). It increases only gradually with the synaptic strength, logarithmically in the case of an exponential connection function and only slightly for a step connection function. It also decreases gradually with a potassium leak conductance that hyperpolarizes RE cells. 7. To reproduce the experimentally measured wavefront velocity of approximately 1 mm/s, together with other in vitro observations, both the RE-to-TC and the TC-to-RE projections in the model should be spatially localized. The sum of the RE-to-TC and the TC-to-RE synaptic footprint lengths should be on the order of 100 microns. (ABSTRACT TRUNCATED AT 250 WORDS)

Action Potentials↗

PubMed2001

Automatic determination of radial basis functions: an immunity-based approach.

The appropriate operation of a radial basis function (RBF) neural network depends mainly upon an adequate choice of the parameters of its basis functions. The simplest approach to train an RBF network is to assume fixed radial basis functions defining the activation of the hidden units. Once the RBF parameters are fixed, the optimal set of output weights can be determined straightforwardly by using a linear least squares algorithm, which generally means reduction in the learning time as compared to the determination of all RBF network parameters using supervised learning. The main drawback of this strategy is the requirement of an efficient algorithm to determine the number, position, and dispersion of the RBFs. The approach proposed here is inspired by models derived from the vertebrate immune system, that will be shown to perform unsupervised cluster analysis. The algorithm is introduced and its performance is compared to that of the random, k-means center selection procedures and other results from the literature. By automatically defining the number of RBF centers, their positions and dispersions, the proposed method leads to parsimonious solutions. Simulation results are reported concerning regression and classification problems.

Algorithms↗

PubMed2004

Physical network models.

We develop a new framework for inferring models of transcriptional regulation. The models, which we call physical network models, are annotated molecular interaction graphs. The attributes in the model correspond to verifiable properties of the underlying biological system such as the existence of protein-protein and protein-DNA interactions, the directionality of signal transduction in protein-protein interactions, as well as signs of the immediate effects of these interactions. Possible configurations of these variables are constrained by the available data sources. Some of the data sources, such as factor-binding data, involve measurements that are directly tied to the variables in the model. Other sources, such as gene knock-outs, are functional in nature and provide only indirect evidence about the variables. We associate each observed knock-out effect in the deletion mutant data with a set of causal paths (molecular cascades) that could in principle explain the effect, resulting in aggregate constraints about the physical variables in the model. The most likely settings of all the variables, specifying the most likely graph annotations, are found by a recursive application of the max-product algorithm. By testing our approach on datasets related to the pheromone response pathway in S. cerevisiae, we demonstrate that the resulting model is consistent with previous studies about the pathway. Moreover, we successfully predict gene knock-out effects with a high degree of accuracy in a cross-validation setting. When applying this approach genome-wide, we extract submodels consistent with previous studies. The approach can be readily extended to other data sources or to facilitate automated experimental design.

Computational Biology↗

PubMed2005

A machine learning evaluation of an artificial immune system.

ARTIS is an artificial immune system framework which contains several adaptive mechanisms. LISYS is a version of ARTIS specialized for the problem of network intrusion detection. The adaptive mechanisms of LISYS are characterized in terms of their machine-learning counterparts, and a series of experiments is described, each of which isolates a different mechanism of LISYS and studies its contribution to the system's overall performance. The experiments were conducted on a new data set, which is more recent and realistic than earlier data sets. The network intrusion detection problem is challenging because it requires one-class learning in an on-line setting with concept drift. The experiments confirm earlier experimental results with LISYS, and they study in detail how LISYS achieves success on the new data set.

Algorithms↗