Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “Neural network”

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.

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

At least 559 records · Page 31Linked to original sources
PubMed2004

Pulse-coupled neural networks for contour and motion matchings.

Two neural networks based on temporal coding are proposed in this paper to perform contour and motion matchings. Both of the proposed networks are three-dimensional (3-D) pulse-coupled neural networks (PCNNs). They are composed of simplified Eckhorn neurons and mimic the structure of the primary visual cortex. The PCNN for contour matching can segment from the background the object with a particular contour, which has been stored as prior knowledge and controls the network activity in the form of spike series; The PCNN for motion matching not only detects the motion in the visual field, but also extracts the object moving in an arbitrarily specified direction. The basic idea of these two models is to encode information into the timing of spikes and later to decode this information through coincidence detectors and synapse delays to realize the knowledge-controlled object matchings. The simulation results demonstrate that the temporal coding and the decoding mechanisms are powerful enough to perform the contour and motion matchings.

Action Potentials↗
PubMed2005

Fast time delay neural networks.

This paper presents a new approach to speed up the operation of time delay neural networks. The entire data are collected together in a long vector and then tested as a one input pattern. The proposed fast time delay neural networks (FTDNNs) use cross correlation in the frequency domain between the tested data and the input weights of neural networks. It is proved mathematically and practically that the number of computation steps required for the presented time delay neural networks is less than that needed by conventional time delay neural networks (CTDNNs). Simulation results using MATLAB confirm the theoretical computations.

Computer Simulation↗
PubMed1999

ANNSyS: an Analog Neural Network Synthesis System.

A synthesis system based on a circuit simulator and a silicon assembler for analog neural networks to be implemented in MOS technology is presented. The system approximates on-chip training of the neural network under consideration and provides the best starting point for 'chip-in-the-loop training'. Behaviour of the analog neural network circuitry is modeled according to its SPICE simulations and those models are used in the initial training of the analog neural networks prior to the fine tuning stage. In this stage, the simulator has been combined with Madaline Rule III for approximating on chip training by software, thus minimizing the effects of circuit nonidealities on neural networks. The circuit simulator partitions the circuit into decoupled blocks which can be simulated separately, with the output of one block being the input for the next one. Finally, the silicon assembler generates the layout for the neural network by reading analog standard cells from a library. The system's performance has been demonstrated by several examples.

Journal Article↗
PubMed1995

The use of neural networks to recognize patterns of human movement: gait patterns.

Artificial neural networks and a statistical method, linear discriminant analysis, were both applied to the recognition of temporal gait parameters associated with altered gait patterns. The duration of the double support and right and left single support phases were measured at seven speeds and three walking conditions. Data from 10 subjects were used to train neural networks, which were then tested using data from 10 other subjects. The overall performance of the networks was at least as high as that of linear discriminant analysis. The relative ease with which neural networks can be set up in a computer, and their discriminatory power, suggests that the technique has a useful role to play in gait analysis. RELEVANCE: The capacity of neural networks to recognize alteration of gait patterns suggests that they might provide an alternative approach for gait assessment. They might be proved to be a useful diagnostic tool.

Journal Article↗
PubMed2006

A fast identification algorithm for Box-Cox transformation based radial basis function neural network.

In this letter, a Box-Cox transformation-based radial basis function (RBF) neural network is introduced using the RBF neural network to represent the transformed system output. Initially a fixed and moderate sized RBF model base is derived based on a rank revealing orthogonal matrix triangularization (QR decomposition). Then a new fast identification algorithm is introduced using Gauss-Newton algorithm to derive the required Box-Cox transformation, based on a maximum likelihood estimator. The main contribution of this letter is to explore the special structure of the proposed RBF neural network for computational efficiency by utilizing the inverse of matrix block decomposition lemma. Finally, the Box-Cox transformation-based RBF neural network, with good generalization and sparsity, is identified based on the derived optimal Box-Cox transformation and a D-optimality-based orthogonal forward regression algorithm. The proposed algorithm and its efficacy are demonstrated with an illustrative example in comparison with support vector machine regression.

Algorithms↗
PubMed2003

Alike performance during nonverbal episodic learning from diversely imprinted neural networks.

Performance on neuropsychological testing permits inferences to be made regarding neural networks required to solve the task. In healthy young human subjects it is common sense that differential performance in cognitive tasks results from recruitment of different neural networks and that alike performance results from recruitment of alike neural networks. It was the goal of the present study to investigate whether these assumptions are also valid in cross-cultural studies. To address this, we used functional MRI during a nonverbal episodic memory task with repeated learning of abstract geometric patterns. Behavioural performance in this task was alike over repeated trials in native Chinese and Caucasian subjects. Given this equivalent performance, the distinct pattern of neuronal activation observed is interpreted as the outcome of different culturally imprinted processing routines. In the 'what' and 'where' framework of visuo-spatial processing initial learning in Chinese subjects activated the dorsal stream for analysis of spatial features whereas Caucasians recruited the ventral stream for object identification. With repeated learning Chinese subjects integrated visuo-spatial processing to object coding and vice versa. Thus, imprints of culture result in activation of distinct neural networks and mandate monitoring of both behavioural performance and neural recruitment in cross-cultural studies of cognition.

Adult↗
PubMed2003

fMRI identifies regional specialization of neural networks for reading in young children.

BACKGROUND: fMRI allows mapping of neural networks underlying cognitive networks during development, but few studies have systematically examined children 7 and younger, in whom language networks may be more diffusely organized than in adults. OBJECTIVE: To identify neural networks during early reading consolidation in young children. METHODS: The authors studied 16 normal, right-handed, native English-speaking children with a mean age of 7.2 years (range 5.8 to 7.9) with fMRI reading paradigms adjusted for reading level. Data were acquired with the echoplanar imaging BOLD technique at 1.5 T. Group data were analyzed with statistical parametric mapping (SPM-99); individual data sets were analyzed with a region of interest approach from individual study t maps (t = 4). The number of activated pixels in brain regions was determined and an asymmetry index (AI) ([L-R]/[L+R]) calculated for each region. RESULTS: In group analysis the authors found prominent activation in left inferior temporal occipital junction and left fusiform gyrus (Brodmann area [BA] 37), middle temporal gyrus (BA 21, 22), middle frontal gyrus (BA 44, 45), and the supplementary motor area. Activation was strongly lateralized in middle frontal gyrus and Wernicke areas (AI 0.54, 0.62). Fourteen subjects had left-sided language lateralization, one was bilateral, and one had poor activation. CONCLUSIONS: The neural networks that process reading are strongly lateralized and regionally specific by age 6 to 7 years. Neural networks in early readers are similar to those in adults.

Brain Mapping↗
PubMed2005

Interpretation of captopril renography using artificial neural networks.

The purpose of this study was to develop a method based on artificial neural networks for interpretation of captopril renography tests for the detection of renovascular hypertension caused by renal artery stenosis and to assess the value of different measurements from the test. A total of 250 99mTc-MAG3 captopril renography tests were used in the study. The material was collected from two different patient groups. One group consisted of 101 patients who also had undergone a renal angiography. The angiographies, which were used as gold standard, showed a significant renal artery stenosis in 53 of the 101 cases. The second group consisted of 149 patients, who's captopril renography tests all were interpreted as not compatible with significant renal artery stenosis by an experienced nuclear medicine physician. Artificial neural networks were trained for the diagnosis of renal artery stenosis using eight measures from each renogram. The neural network was then evaluated in separate test groups using an eightfold cross validation procedure. The performance of the neural networks, measured as the area under the receiver operating characteristic curve, was 0.93. The sensitivity was 91% at a specificity of 90%. The lowest performance was found for the network trained without use of a parenchymal transit measure, indicating the importance of this feature. Artificial neural networks can be trained to interpret captopril renography tests for detection of renovascular hypertension caused by renal artery stenosis. The result almost equals that of human experts shown in previous studies.

Angiotensin-Converting Enzyme Inhibitors↗
PubMedDate not supplied

Extracting the principal behavior of a probabilistic supervisor through neural networks ensemble.

In this paper, we propose a model of a neural network ensemble that can be trained with a supervisor having two kinds of input-output functions where the occurrence probability of each function is not even. This condition can be likened to a learning condition, in which the learning data are hampered by noise. In this case, the neural network has the impression that the learning supervisor (object) has a probabilistic behavior in which the supervisor generates correct learning data most of the time but occasionally generates erroneous ones. The objective is to train the neural network to approximate the greatest distributed input-output relation, which can be considered to be the principal nature of the supervisor, so that we can obtain a neural network that is able, to some extent, to suppress the ill effect of erroneous data encountered during the learning process.

Algorithms↗
PubMed1997

[Intraoperative EEG monitoring using a neural network].

OBJECTIVE: To introduce a new EEG parameter for monitoring a patient's cerebral status during anaesthesia. METHOD: EEG epochs (channel C3 P3, duration 30 see per epoch) yield patterns used as the training input of a self-organising neural network (neural gas algorithm). Each pattern contains spectral components. An additional "suppression parameter" reflects the proportion of flat curves during an EEG epoch and reduces the shortcomings of spectral analysis. The enhanced pattern vector enables the recognition of burst-suppression periods representing depressed cerebral activity during anaesthesia. Following training with 25549 EEG epochs ¿recorded in 196 consenting patients in the period beginning 5 minutes before induction of anaesthesia to extubation¿ the network knows 125 basic patterns (neural clusters). The neural clusters are responsible for different stages of anaesthesia: some neurons are activated by EEG epochs of awake patients, others are stimulated by deep stages. The homogeneity of the EEG epochs of cluster is checked. The portion of the EEG epochs not recognised by the network (7.5%) contains heterogeneous and sporadic patterns (mainly outlines caused by artefacts). RESULTS: In comparison with commonly used variables of anaesthesiological EEG monitoring (spectral edge or median frequency) neural discriminant analysis achieves better discrimination between awake and deeply anaesthetised stages (reclassification of 96% vs. 70%). On the basis of 56 complete sequences of patterns (from the beginning of the infusion of anaesthetics to the occurrence of burst suppression) a trend value of between 0.0 and 100.0 is assigned to each neural cluster. In contrast to existing methods, induction of anaesthesia causes a strictly linear increase in the electroencephalographic trend. CONCLUSIONS: Detailed neural cluster and discriminant analysis on the basis of the model described leads to an improved EEG parameter which better reflects the hypnotic effects of anaesthetic agents and arousal reactions caused by pain stimuli. Misclassifications of awake and anaesthetised stages are reduced. The neural network learns to recognize the complex changes in EEG patterns during induction of anaesthesia with different agents.

Adult↗
PubMed2002

Dynamic stability conditions for Lotka-Volterra recurrent neural networks with delays.

The Lotka-Volterra model of neural networks, derived from the membrane dynamics of competing neurons, have found successful applications in many "winner-take-all" types of problems. This paper studies the dynamic stability properties of general Lotka-Volterra recurrent neural networks with delays. Conditions for nondivergence of the neural networks are derived. These conditions are based on local inhibition of networks, thereby allowing these networks to possess a multistability property. Multistability is a necessary property of a network that will enable important neural computations such as those governing the decision making process. Under these nondivergence conditions, a compact set that globally attracts all the trajectories of a network can be computed explicitly. If the connection weight matrix of a network is symmetric in some sense, and the delays of the network are in L2 space, we can prove that the network will have the property of complete stability.

Journal Article↗
PubMed1997

Relating formulation variables to in vitro dissolution using an artificial neural network.

The purpose of this paper was to investigate the effect of several experimental variables on the ability of a neural network to predict in vitro dissolution rate as a function of product formulation changes. Neural network software was trained with sets of hypothetical and experimental data consisting of 4-15 formulations with known in vitro drug dissolution profiles and the ability of the trained model to recognize patterns was validated against similar formations not used to train the neural network. The effect of selected variables, e.g., number of hidden-layer nodes and iterations, as well as the use of replicate or mean data on the accuracy of the predictions was investigated. The importance of optimizing the number of hidden-layer nodes and iterations was demonstrated. The prediction error increased for validation data sets that were outside the range of the training data set. Accurate predictions were obtained with as few as four formulations in the training set, provided the formulations were carefully chosen, and the number of formulation variables were small. Also, limiting the validation set to one formulation was not sufficient to validate the neural network model. Increasing the size of the training set, or replication of the input and output data, also provided more accurate predictions. The neural network accurately predicted in vitro drug release provided the neural network variables were optimized, and the training and validation data sets were appropriately selected.

Acetaminophen↗
PubMed2004

A recurrent neural network with exponential convergence for solving convex quadratic program and related linear piecewise equations.

This paper presents a recurrent neural network for solving strict convex quadratic programming problems and related linear piecewise equations. Compared with the existing neural networks for quadratic program, the proposed neural network has a one-layer structure with a low model complexity. Moreover, the proposed neural network is shown to have a finite-time convergence and exponential convergence. Illustrative examples further show the good performance of the proposed neural network in real-time applications.

Artificial Intelligence↗
PubMed2000

Neural spike sorting under nearly 0-dB signal-to-noise ratio using nonlinear energy operator and artificial neural-network classifier.

We report a result on neural spike sorting under conditions where the signal-to-noise ratio is very low. The use of nonlinear energy operator enables the detection of an action potential, even when the SNR is so poor that a typical amplitude thresholding method cannot be applied. The superior detection ability facilitates the collection of a training set under lower SNR than that of the methods which employ simple amplitude thresholding. Thus, the statistical characteristics of the input vectors can be better represented in the neural-network classifier. The trained neural-network classifiers yield the correct classification ratio higher than 90% when the SNR is as low as 1.2 (0.8 dB) when applied to data obtained from extracellular recording from Aplysia abdominal ganglia using a semiconductor microelectrode array.

Action Potentials↗
PubMed2005

Analysis of the applicability of artificial neural networks for studying blood plasma: determination of magnesium ion concentration as a case study.

Artificial neural networks are suggested for use in predicting metal ion concentration in human blood plasma. Simulated and available experimental data are used to train the artificial neural network. Particularly, using 850 simulated samples, the network predicted the magnesium-free ion concentration with an average error smaller than 1%. Clinical data recently reported for 20 patients were considered and the artificial neural network predicted the concentration of free magnesium ion with an average error of about 6%. Overall, the approach of using artificial neural networks as an alternative or complementary strategy to deal with the analysis of human blood plasma can be useful for clinical diagnostics, if there is sufficient data to train the artificial neural network.

Humans↗
PubMed1999

Prediction of oculocardiac reflex in strabismus surgery using neural networks.

Successfully predicting an oculocardiac reflex (OCR) is difficult to achieve despite various proposed maneuvers. The aim of this study was to test the models built up by neural networks to predict the occurrence of OCR during strabismus surgery in children. Premedication was not given. Atropine 0.01 mg/kg was medicated just before induction. Induction was performed with fentanyl or ketorolac, followed by propofol. Atracurium or vecuronium was given for intubation. Anesthesia was maintained with O2-N2O with continuous propofol infusion. Chi-square test was performed for induction agents, gender, weight, muscle blockade, repaired muscle, number of repaired muscles, duration of operation to detect any association between the occurrence of OCR and to develop the model of neural networks. The multi-layer perceptron, radial basis function and Bayesian backpropagation network were tested. The occurrence of OCR was significantly associated with gender and repaired muscle (p < 0.05). Gender, repaired muscle and age were considered as input for the multi-layer perceptron, radial basis function and Bayesian backpropagation network. Three neural networks had predicted the same correction rate in the occurrence of OCR as being 87.5% overall among 16 patients' records tested. These models are conceptually different in predicting compared to conventional maneuvers, and have the advantage of testing individually and foretelling the propensity. By comparison neural networks use grouped experiential data and predict OCR by the learning rule. Neural networks require a relatively abundant number of experienced and homogenous patients' records to establish an accurate model. The multi-layer perceptron, radial basis function and Bayesian backpropagation modeling network may be an alternative way, and preferable to vagal tone maneuvers if the associated relationships to the occurrence of OCR are more clearly defined.

Adolescent↗
PubMed1997

Removing irrelevant features in neural network classification using evolutionary computations.

Evolutionary artificial neural networks (EANN) are a new paradigm that refers to a special class of artificial neural networks (ANN) in which evolution is another fundamental form of adaptation in addition to learning. Evolution can be introduced at various levels of ANN. It can be used to evolve weights, architectures and learning parameters. Evolutionary computations are population-based search methods that have shown promise in many similarly complex tasks. This paper presents an application of evolutionary programming for simultaneously inducing the input structure and weights evolving for multilayer feed-forward perceptrons (MLP) with standard sigmoidal activation function.

Algorithms↗
PubMed1994

Back-propagation and counter-propagation neural networks for phylogenetic classification of ribosomal RNA sequences.

A neural network system has been developed for rapid and accurate classification of ribosomal RNA sequences according to phylogenetic relationship. The molecular sequences are encoded into neural input vectors using an n-gram hashing method. A SVD (singular value decomposition) method is used to compress and reduce the size of long and sparse n-gram input vectors. The neural networks used are three-layered, feed-forward networks that employ supervised learning paradigms, including the back-propagation algorithm and a modified counter-propagation algorithm. A pedagogical pattern selection strategy is used to reduce the training time. After trained with ribosomal RNA sequences of the RDP (Ribosomal Database Project) database, the system can classify query sequences into more than one hundred phylogenetic classes with a 100% accuracy at a rate of less than 0.3 CPU second per sequence on a workstation. When compared to other sequence similarity search methods, including Similarity Rank, Blast and Fasta, the neural network method has a higher classification accuracy at a speed of about an order of magnitude faster. The software tool will be made available to the biology community, and the system may be extended into a gene identification system for classifying indiscriminately sequenced DNA fragments.

Algorithms↗