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 1,477 records · Page 82Linked to original sources

PubMed2005

An incremental regression method for graph structured data.

In this paper, we consider learning problems defined on graph-structured data. We propose an incremental supervised learning algorithm for network-based estimators using diffusion kernels. Diffusion kernel nodes are iteratively added in the training process. For each new node added, the kernel function center and the output connection weight are decided according to an empirical risk driven rule based on an extended chained version of the Nadaraja-Watson estimator. Then the diffusion parameters are determined by a genetic-like optimization technique.

Algorithms↗

PubMed1997

Reasoning requirements for diagnosis of heart disease.

Over the past dozen years, the Heart Disease Program (HDP) has been developed to assist physicians in reasoning about cardiovascular disorders. Driven by several evaluations, the inference mechanism has progressed from a logic based model, to a Bayesian Probability Network (BPN) and finally a pseudo-Bayesian network with temporal and severity reasoning. Though aspects of cardiovascular reasoning are handled well by BPNs, temporal reasoning, homeostatic feedback mechanisms and effects of disease severities require additional inference strategies. This article discusses how these reasoning problems are handled, and deals with closely linked issues in building the user interface to collect detailed cardiovascular data and provide clear explanations of diagnoses.

Artificial Intelligence↗

PubMed1999

Alpha rhythm emerges from large-scale networks of realistically coupled multicompartmental model cortical neurons.

Conical pyramidal and stellate neurons were simulated using the GENESIS simulation package. Model neurons were leaky integrate-and-fire and consisted of from four to nine passive compartments. Neurophysiological measurements, based on single-cell recordings and patch-clamp experiments, provided estimations for the simulation of cortical neurons: transmitter-activated conductances, passive membrane time constants and axonal delays. Network connectivity was generated using a previously described probabilistic scheme based on known cortical histology, in which the probability of connections forming between one neuron and another fell off monotonically with increasing inter-cellular separation. Simulations of up to 6400 cortical neurons, approaching the scale of an individual cortical column, confirmed previous findings with smaller networks. Limit-cycle behaviour emerged in the network, in the frequency in the range of the mammalian alpha and beta rhythms (8-20 Hz). Contrary to expectation, near-linear relationships were found between the mean soma membrane potential and and neuronal firing probability. Some of the implications for cortical information processing, in particular the dynamical interactions between the neuronal and larger scales, are discussed.

Alpha Rhythm↗

PubMed1995

An artificial neural network approach to ERP classification.

Artificial neural network-based approaches were developed to classify event-related potential (ERP) waveforms. The networks utilized scalp-recorded ERP returns from six electrode sites. These ERPs were evoked as one individual responded to a series of auditorily presented object names while viewing various objects on a computer screen. The ERPs at the electrode sites were classified as a match decision or a no-match decision. A three-layer backpropagation neural network model was selected to formulate a global and a local classification approach. The backpropagation network in the global approach was designed to operate on a single ERP response which was the average of the ERP responses generated at the six electrode sites. The local ERP classification system consisted of six three-layer backpropagation networks. Each network was designed to operate on the ERPs generated at a single electrode site. A small data base consisting of eight match and eight no-match ERP responses was used to train and test the networks in a variety of ways. The results obtained clearly show that the neural network-based classifiers are able to discriminate with a high degree of accuracy between match and no-match conditions in ERP waveforms.

Brain↗

PubMed2003

Neuronal avalanches in neocortical circuits.

Networks of living neurons exhibit diverse patterns of activity, including oscillations, synchrony, and waves. Recent work in physics has shown yet another mode of activity in systems composed of many nonlinear units interacting locally. For example, avalanches, earthquakes, and forest fires all propagate in systems organized into a critical state in which event sizes show no characteristic scale and are described by power laws. We hypothesized that a similar mode of activity with complex emergent properties could exist in networks of cortical neurons. We investigated this issue in mature organotypic cultures and acute slices of rat cortex by recording spontaneous local field potentials continuously using a 60 channel multielectrode array. Here, we show that propagation of spontaneous activity in cortical networks is described by equations that govern avalanches. As predicted by theory for a critical branching process, the propagation obeys a power law with an exponent of -3/2 for event sizes, with a branching parameter close to the critical value of 1. Simulations show that a branching parameter at this value optimizes information transmission in feedforward networks, while preventing runaway network excitation. Our findings suggest that "neuronal avalanches" may be a generic property of cortical networks, and represent a mode of activity that differs profoundly from oscillatory, synchronized, or wave-like network states. In the critical state, the network may satisfy the competing demands of information transmission and network stability.

Animals↗

PubMed2003

A two-stage unsupervised learning algorithm reproduces multisensory enhancement in a neural network model of the corticotectal system.

Multisensory enhancement (MSE) is the augmentation of the response to sensory stimulation of one modality by stimulation of a different modality. It has been described for multisensory neurons in the deep superior colliculus (DSC) of mammals, which function to detect, and direct orienting movements toward, the sources of stimulation (targets). MSE would seem to improve the ability of DSC neurons to detect targets, but many mammalian DSC neurons are unimodal. MSE requires descending input to DSC from certain regions of parietal cortex. Paradoxically, the descending projections necessary for MSE originate from unimodal cortical neurons. MSE, and the puzzling findings associated with it, can be simulated using a model of the corticotectal system. In the model, a network of DSC units receives primary sensory input that can be augmented by modulatory cortical input. Connection weights from primary and modulatory inputs are trained in stages one (Hebb) and two (Hebb-anti-Hebb), respectively, of an unsupervised two-stage algorithm. Two-stage training causes DSC units to extract information concerning simulated targets from their inputs. It also causes the DSC to develop a mixture of unimodal and multisensory units. The percentage of DSC multisensory units is determined by the proportion of cross-modal targets and by primary input ambiguity. Multisensory DSC units develop MSE, which depends on unimodal modulatory connections. Removal of the modulatory influence greatly reduces MSE but has little effect on DSC unit responses to stimuli of a single modality. The correspondence between model and data suggests that two-stage training captures important features of self-organization in the real corticotectal system.

Algorithms↗

PubMed2005

Sequential information processing using time-delay connections in ontogenic CALM networks.

In this paper, a variant of the categorization-and-learning-module (CALM) network is presented that is not only capable of categorizing sequential information with feedback, but can adapt its resources to the current training set. In other words, the modules of the network may grow or shrink depending on the complexity of the presented sequence-set. In the original CALM algorithm, modules did not have access to activations from earlier stimulus presentations. To bypass this limitation, we introduced time-delay connections in CALM. These connections allow for a delayed propagation of activation, such that information at a given time will be available to a module at a later timestep. In addition, modules can autonomously add and remove resources depending on the structure and complexity of the task domain. The performance of this ontogenic CALM network with time-delay connections is demonstrated and analyzed using a sample set of overlapping sequences from an existing problem domain.

Algorithms↗

PubMed2004

Discrimination networks for maximum selection.

We construct a novel discrimination network using differentiating units for maximum selection. In contrast to traditional competitive architectures like MAXNET the discrimination network does not only signal the winning unit, but also provides information about its evidence. In particular, we show that a discrimination network converges to a stable state within finite time and derive three characteristics: intensity normalization (P1), contrast enhancement (P2), and evidential response (P3). In order to improve the accuracy of the evidential response we incorporate distributed redundancy into the network. This leads to a system which is not only robust against failure of single units and noisy data, but also enables us to sharpen the focus on the problem given in terms of a more accurate evidential response. The proposed discrimination network can be regarded as a connectionist model for competitive learning by evidence.

Computer Simulation↗

PubMed2003

An efficient approximation algorithm for finding a maximum clique using Hopfield network learning.

In this article, we present a solution to the maximum clique problem using a gradient-ascent learning algorithm of the Hopfield neural network. This method provides a near-optimum parallel algorithm for finding a maximum clique. To do this, we use the Hopfield neural network to generate a near-maximum clique and then modify weights in a gradient-ascent direction to allow the network to escape from the state of near-maximum clique to maximum clique or better. The proposed parallel algorithm is tested on two types of random graphs and some benchmark graphs from the Center for Discrete Mathematics and Theoretical Computer Science (DIMACS). The simulation results show that the proposed learning algorithm can find good solutions in reasonable computation time.

Algorithms↗

PubMed2005

Emotion recognition in human-computer interaction.

In this paper, we outline the approach we have developed to construct an emotion-recognising system. It is based on guidance from psychological studies of emotion, as well as from the nature of emotion in its interaction with attention. A neural network architecture is constructed to be able to handle the fusion of different modalities (facial features, prosody and lexical content in speech). Results from the network are given and their implications discussed, as are implications for future direction for the research.

Attention↗

PubMed2006

Oscillating neurons in the cochlear nucleus: II. Simulation results.

A computer model of sustained chopper neurons in the ventral cochlear nucleus is presented and investigated. In the companion paper, the underlying neurophysiological and neuroanatomical data are demonstrated. To explain the preference of chopper neurons for oscillations with periods which are multiples of a 0.4 ms synaptic delay, we suggest a model of circularly connected chopper neurons. In order to simulate chopper neurons within a physiological dynamic range for periodicity encoding, it is necessary to assume that they receive an input from onset neurons. Our computer analysis of the resulting simple neuronal network shows that it can produce stable oscillations. The chopping can be triggered by an amplitude-modulated signal (AM). The dynamic range and the synchronous response of the simulated chopper neurons to AM are enhanced significantly by an additional input from onset neurons. Physiological properties of chopper neurons in the cat, such as mean, standard deviation, and coefficient of variation of the interspike interval are matched precisely by our simulations.

Acoustic Stimulation↗

PubMed2005

Recent advances in gene expression data clustering: a case study with comparative results.

Several advanced techniques have been proposed for data clustering and many of them have been applied to gene expression data, with partial success. The high dimensionality and the multitude of admissible perspectives for data analysis of gene expression require additional computational resources, such as hierarchical structures and dynamic allocation of resources. We present an immune-inspired hierarchical clustering device, called hierarchical artificial immune network (HaiNet), especially devoted to the analysis of gene expression data. This technique was applied to a newly generated data set, involving maize plants exposed to different aluminum concentrations. The performance of the algorithm was compared with that of a self-organizing map, which is commonly adopted to deal with gene expression data sets. More consistent and informative results were obtained with HaiNet.

Algorithms↗

PubMed2004

Autoassociator networks: insights into infant cognition.

This paper presents autoassociator neural networks. A first section reviews the architecture of these models, common learning rules, and presents sample simulations to illustrate their abilities. In a second section, the ability of these models to account for learning phenomena such as habituation is reviewed. The contribution of these networks to discussions about infant cognition is highlighted. A new, modular approach is presented in a third section. In the discussion, a role for these learning models in a broader developmental framework is proposed.

Artificial Intelligence↗

PubMed1996

Kohonen map as a visualization tool for the analysis of protein sequences: multiple alignments, domains and segments of secondary structures.

The method of Kohonen maps, a special form of neural networks, was applied as a visualization tool for the analysis of protein sequence similarity. The procedure converts sequence (domains, aligned sequences, segments of secondary structure) into a characteristic signal matrix. This conversion depends on the property or replacement score vector selected by the user. Similar sequences have small distance in the signal space. The trained Kohonen network is functionally equivalent to an unsupervised non-linear cluster analyzer. Protein families, or aligned sequences, or segments of similar secondary structure, aggregate as clusters, and their proximity may be inspected on a color screen or on paper. Pull-down menus permit access to background information in the established text-oriented way.

Amino Acid Sequence↗

PubMed1995

Mathematical and connectionist models of human memory: a comparison.

Recent convolution-based models of human memory (e.g. Lewandowsky & Murdock, 1989), have accounted for a wide range of data. However such models require the relevant mathematical operations to be provided to the network. Connectionist models, in contrast, have generally addressed different data, and not all architectures are appropriate for modelling single-trial learning. Furthermore, they tend to exhibit catastrophic interference in multiple list learning. In this paper we compare the ability of convolution-based models and DARNET (Developmental Associative Recall NETwork), to account for human memory data. DARNET is a connectionist approach to human memory in which the system gradually learns to associate vectors, in one trial, into a memory trace vector. Either of the vectors can than be retrieved. It is shown that the new associative mechanism can be used to account for a wide range of relevant experimental data as successfully as can convolution-based models with the same higher-level architectures. Limitations of the models are also addressed.

Computer Simulation↗

PubMed1998

Computational model of obsessive-compulsive disorder: examination of etiologic hypothesis and treatment strategies.

Research has shown that obsessive compulsive disorder (OCD) is related to structural and functional abnormalities in the brain, and several authors have organized these findings into theories of OCD neuropsychiatric dysfunction. In this paper, these theories were used to develop a neural network model of OCD. OCD symptoms were hypothesized to result from a hyperactive orbitofrontal-striato-thalamic-orbitofrontal neural loop. The network was constructed and trained with a backpropagation algorithm, and it was then used to assess etiologic theories of OCD (e.g., basal ganglia dysfunction, inadequate dopaminergic inhibitory influence on basal ganglia and excessive input from the limbic system). The network was also observed in analogues of the treatment of OCD with serotonergic medications and behavior therapy. Results show that a) the network behaved both normally and abnormally, depending on what combinations of perceptual, motivational, and neurochemical inputs were presented to it; b) several etiologic mechanisms produced changes in the networks' behaviors similar to patients' subjective experiences of OCD symptoms; and c) different treatment strategies, both those modeled as pharmacologic and behavioral therapies, produced reductions in simulated OCD symptoms.

Algorithms↗

PubMed2006

Dynamical network interactions in distributed control of robots.

In this paper the dynamical network model of the interactions within a group of mobile robots is investigated and proposed as a possible strategy for controlling the robots without central coordination. Motivated by the results of the analysis of our simple model, we show that the system performance in the presence of noise can be improved by including long-range connections between the robots. Finally, a suitable strategy based on this model to control exploration and transport is introduced.

Algorithms↗

PubMed2004

Neural network approach to the segmentation and classification of dynamic magnetic resonance images of the breast: comparison with empiric and quantitative kinetic parameters.

RATIONALE AND OBJECTIVE: An artificial neural network (ANN)-based segmentation method was developed for dynamic contrast-enhanced magnetic resonance (MR) imaging of the breast and compared with quantitative and empiric parameter mapping techniques. MATERIALS AND METHODS: The study population was composed of 10 patients with seven malignant and three benign lesions undergoing dynamic MR imaging of the breast. All lesions were biopsied or surgically excised, and examined by means of histopathology. A T1-weighted 3D FLASH (fast low angle shot sequence) was acquired before and seven times after the intravenous administration of gadopentetate dimeglumine at a dose of 0.1 mmol/kg body weight. Motion artifacts on MR images were eliminated by voxel-based affine and nonrigid registration techniques. A two-layered feed-forward back-propagation network was created for pixel-by-pixel classification of signal intensity-time curves into benign/malignant tissue types. ANN output was statistically compared with percent-enhancement (E), signal enhancement ratio (SER), time-to-peak, subtracted signal intensity (SUB), pharmacokinetic parameter rate constant (k(ep)), and correlation coefficient to a predefined reference washout curve. RESULTS: ANN was successfully applied to the classification of breast MR images identifying structures with benign or malignant enhancement kinetics. Correlation coefficient (logistic regression, odds ratio [OR] = 12.9; 95% CI: 7.7-21.8), k(ep) (OR = 1.8; 95% CI: 1.2-2.6), and time-to-peak (OR = 0.45; 95% CI: 0.3-0.7) were independently associated to ANN output classes. SER, E, and SUB were nonsignificant covariates. CONCLUSION: ANN is capable of classifying breast lesions on MR images. Mapping correlation coefficient, k(ep) and time-to-peak showed the highest association with the ANN result.

Adult↗