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 127 records · Page 7Linked to original sources

PubMedDate not supplied

The Wisconsin Card Sorting Test: theoretical analysis and modeling in a neuronal network.

Neuropsychologists commonly use the Wisconsin Card Sorting Test as a test of the integrity of frontal lobe functions. However, an account of its range of validity and of the neuronal mechanisms involved is lacking. We analyze the test at 3 different levels. First, the different versions of the test are described, and the results obtained with normal subjects and brain-lesioned patients are reviewed. Second, a computational analysis is used to reveal what algorithms may pass the test, and to predict their respective performances. At this stage, 3 cognitive components are isolated that may critically contribute to performance: the ability to change the current rule when negative reward occurs, the capacity to memorize previously tested rules in order to avoid testing them twice, and the possibility of rejecting some rules a priori by reasoning. Third, a model neuronal network embodying these 3 components is described. The coding units are clusters of neurons organized in layers, or assemblies. A sensorimotor loop enables the network to sort the input cards according to several criteria (color, form, etc.). A higher-level assembly of rule-coding clusters codes for the currently tested rule, which shifts when negative reward is received. Internal testing of the possible rules, analogous to a reasoning process, also occurs, by means of an endogenous auto-evaluation loop. When lesioned, the model reproduces the behavior of frontal lobe patients. Plausible biological or molecular implementations are presented for several of its components.

Frontal Lobe↗

PubMed1990

Predicting surface exposure of amino acids from protein sequence.

The amino acid residues on a protein surface play a key role in interaction with other molecules, determined many physical properties, and constrain the structure of the folded protein. A database of monomeric protein crystal structures was used to teach computer-simulated neural networks rules for predicting surface exposure from local sequence. These trained networks are able to correctly predict surface exposure for 72% of residues in a testing set using a binary model, (buried/exposed) and for 54% of residues using a ternary model (buried/intermediate/exposed). In the ternary model, only 11% of the exposed residues are predicted as buried and only 5% of the buried residues are predicted as exposed. Also, since the networks are able to predict exposure with a quantitative confidence estimate, it is possible to assign exposure for over half of the residues in a binary model with greater than 80% accuracy. Even more accurate predictions are obtained by making a consensus prediction of exposure for a homologous family. The effect of the local environment of an amino acid on its accessibility, though smaller than expected, is significant and accounts for the higher success rate of prediction than obtained with previously used criteria. In the absence of a three-dimensional structure, the ability to predict surface accessibility of amino acids directly from the sequence is a valuable tool in choosing sites of chemical modification or specific mutations and in studies of molecular interaction.

Amino Acid Sequence↗

PubMed1992

An investigation of tumor 1H nuclear magnetic resonance spectra by the application of chemometric techniques.

1H nuclear magnetic resonance (NMR) spectra of tumors and normal tissue include signals from all hydrogen-containing metabolites and can therefore be considered multicomponent multivariate mixtures. We have obtained 1H spectra from perchloric acid extracts of three normal tissues (liver, kidney, and spleen) and five rat tumors (GH3 prolactinoma, Morris hepatomas 7777 and 9618a, LBDS1 fibrosarcoma, and Walker 256 carcinosarcoma). We have applied several different chemometric methods to analyze the data. First, we used principal component analysis, cluster analysis, and an optimized artificial neural network to develop a classification rule from a training set of samples of known origin or class. The classification rule was then assessed using a set of unknown samples. We were able to successfully determine the class of each unknown sample. Second, we used the chemometric techniques of factor analysis followed by target testing to investigate the underlying biochemical differences that are detected between the classes of samples.

Adenocarcinoma↗

PubMed1991

Organizational levels of the cerebral cortex: an integrated model.

We propose a theoretical model of the cerebral cortex which is based on its cellular components and integrates its different levels of organization: (1) cells have general adaptive and memorization properties; (2) cortical columns are repetitive interneuronal circuits which determine an adaptive processing specific to the cerebral cortex; (3) cortical maps effect selective combinations which are very efficient to learn basic behavioural adaptations such as invariant recognition of forms, visually-guided hand movements, or execution of structured motor programs; (4) the network between cortical areas has a global architecture which integrates successive learning experiences into coherent functions such as the human language.

Cerebral Cortex↗

PubMed1992

Simulating vestibular compensation using recurrent back-propagation.

Vestibular compensation is simulated as learning in a dynamic neural network model of the horizontal vestibulo-ocular reflex (VOR). The bilateral, three-layered VOR model consists of nonlinear units representing horizontal canal afferents, vestibular nuclei (VN) neurons and eye muscle motoneurons. Dynamic processing takes place via commissural connections that link the VN bilaterally. The intact network is trained, using recurrent back-propagation, to produce the VOR with velocity storage integration. Compensation is simulated by removing vestibular afferent input from one side and retraining the network. The time course of simulated compensation matches that observed experimentally. The behavior of model VN neurons in the compensated network also matches real data, but only if connections at the motoneurons, as well as at the VN, are allowed to be plastic. The dynamic properties of real VN neurons in compensated and normal animals are found to differ when tested with sinusoidal but not with step stimuli. The model reproduces these conflicting data, and suggests that the disagreement may be due to VN neuron nonlinearity.

Algorithms↗

PubMed1991

Altering the synchrony of stimulus trace processes: tests of a neural-network model.

A previously described neural-network model (Desmond 1991; Desmond and Moore 1988; Moore et al. 1989) predicts that both CS-onset-evoked and CS-offset-evoked stimulus trace processes acquire associative strength during classical conditioning, and that CR waveforms can be altered by manipulating the time at which the processes are activated. In a trace conditioning paradigm, where CS offset precedes US onset, the model predicts that onset and offset traces act in synchrony to generate unimodal CR waveforms. However, if the CS duration is subsequently lengthened on CS-alone probe trials, the model predicts that onset and offset traces will asynchronously contribute to CR output and bimodal CRs will be generated. In a delay conditioning paradigm, in which US onset occurs prior to CS offset, the model predicts that only the onset process will gain associative strength, and hence, only unimodal CRs will occur. Using the rabbit conditioned nictitating membrane response preparation, we found experimental support for these predictions.

Analysis of Variance↗

PubMed1992

Evidence of hyperplanes in the genetic learning of neural networks.

Genetic Algorithms have been successfully applied to the learning process of neural networks simulating artificial life. In previous research we compared mutation and crossover as genetic operators on neural networks directly encoded as real vectors (Manczer and Parisi 1990). With reference to crossover we were actually testing the building blocks hypothesis, as the effectiveness of recombination relies on the validity of such hypothesis. Even with the real genotype used, it was found that the average fitness of the population of neural networks is optimized much more quickly by crossover than it is by mutation. This indicated that the intrinsic parallelism of crossover is not reduced by the high cardinality, as seems reasonable and has indeed been suggested in GA theory (Antonisse 1989). In this paper we first summarize such findings and then propose an interpretation in terms of the spatial correlation of the fitness function with respect to the metric defined by the average steps of the genetic operators. Some numerical evidence of such interpretation is given, showing that the fitness surface appears smoother to crossover than it does to mutation. This confirms indirectly that crossover moves along privileged directions, and at the same time provides a geometric rationale for hyperplanes.

Algorithms↗

PubMed1992

Information maintenance and statistical dependence reduction in simple neural networks.

This study compares the ability of excitatory, feed-forward neural networks to construct good transformations on their inputs. The quality of such a transformation is judged by the minimization of two information measures: the information loss of the transformation and the statistical dependency of the output. The networks that are compared differ from each other in the parametric properties of their neurons and in their connectivity. The particular network parameters studied are output firing threshold, synaptic connectivity, and associative modification of connection weights. The network parameters that most directly affect firing levels are threshold and connectivity. Networks incorporating neurons with dynamic threshold adjustment produce better transformations. When firing threshold is optimized, sparser synaptic connectivity produces a better transformation than denser connectivity. Associative modification of synaptic weights confers only a slight advantage in the construction of optimal transformations. Additionally, our research shows that some environments are better suited than others for recording. Specifically, input environments high in statistical dependence, i.e. those environments most in need of recoding, are more likely to undergo successful transformations.

Action Potentials↗

PubMed1992

A unified neural network [corrected] model of spatiotemporal processing in X and Y retinal ganglion cells. I. Analytical results.

This work presents unified analyses of spatial and temporal visual information processing in a feed-forward network of neurons that obey membrane, or shunting equations. The feed-forward shunting network possesses properties that make it well suited for processing of static, spatial information. However, it is shown here that those same properties of the shunting network that lead to good spatial processing imply poor temporal processing characteristics. This article presents an extension of the feed-forward shunting network model that solves this problem by means of preprocessing layers. The anatomical interpretation of the resulting model is structurally analogous to recently discovered data on a retinal circuit connecting cones to retinal ganglion cells through pairs of push-pull bipolar cells. Mathematical analysis of the lumped model leads to the hypothesis that X and Y retinal ganglion cells may consist of a single mechanism acting in different parameter ranges. This hypothesis is confirmed in the companion article, wherein the model--in conjunction with a nonlinear temporal adaptation mechanism--is used to reproduce experimental data of both X and Y cells by simple changes in morphological and physiological parameters.

Mathematics↗

PubMed1992

A unified neural network model of spatiotemporal processing in X and Y retinal ganglion cells. II. Temporal adaptation and simulation of experimental data.

This article makes use of a push-pull shunting network, which was introduced in the companion article, to model certain properties of X and Y retinal ganglion cells. Input to the push-pull network is preprocessed by a nonlinear mechanism for temporal adaptation, which is ascribed here to photoreceptor dynamics. The complete circuit is used to show that a simple change in receptive field morphology within a single model equation can change the network's response characteristics to closely resemble those of either X or Y cells. Specifically, an increase in width of the receptive field center mechanism is sufficient to account for generation of on-off (Y-like) instead of null (X-like) responses to modulated gratings. In agreement with experimental data, the Y cell on-off response is independent of spatial phase. Also, the model accurately predicts that on-off responses can be observed in X cells for particular stimulus configurations. Taken together, the results show how the retina combines individually inadequate modules to efficiently handle the tasks required for accurate spatial and temporal visual information processing. The model is also able to clarify a number of controversial experimental findings on the nature of spatiotemporal visual processing in the retina.

Adaptation, Physiological↗

PubMed1992

"Live" neuron and optimal learning rule.

A concept of the live unit as an automatic regulation system with a few admissible states areas in the space of states is considered. Energetic profit of oscillatory behavior consisting in the consecutive transitions of system from one admissible states area to another is shown. It is stated, that external disturbances cause the energy consumption of oscillatory system to decrease. On the basis of this concept and some neurophysiological data, the "live" energy-consuming nonlinear three-state neuron model is proposed and the existence of energy optimal generation frequency v(opt) is proved. For the realization of tendency to v(opt) the optimal learning rule is proposed, which provides unsupervised learning and interlinked short-term and long-term memories with forgetting. The model proposed explains the genesis of neural network, is promising in the sense of network self-organization and allows to solve the problem of internal activity in the researches on artificial intelligence.

Biological Clocks↗

PubMed1992

Self-organization of day cycle and hierarchical associative memory in "live" neural network.

The "live" neural network model is proposed on the basis of "live" neuron model and optimal learning rule. By means of numerical simulation the initial stages of neural network self-organization have been shown: (1) the formation of two activity forms, which are identified with sleep and awaking, and (2) the self-organization of hierarchical associative memory when feeding a receptor excitation to the neural network. The energetic profit of self-organization is demonstrated. The formation of neural ensembles, playing the role of generalized neurons, is obtained.

Biological Clocks↗

PubMed1992

On the performance of single-layered neural networks.

This paper studies the performance of single-layered neural networks. This study begins with the performance of single-layered neural networks trained using the outer-product rule. The outer-product rule is a suboptimal learning scheme, resulting under certain assumptions from optimal least-squares training of single-layered neural networks with respect to their analog output. Extensive analysis reveals the improvement on the network performance caused by its optimal least-squares training. The effect of the training scheme on the performance of single-layered neural networks with binary output is exhibited by experimentally comparing the performance of single-layered neural networks trained with respect to their analog and binary output.

Mathematics↗

PubMed1992

Bifurcation analysis of a neural network model.

This paper describes the analysis of the well known neural network model by Wilson and Cowan. The neural network is modeled by a system of two ordinary differential equations that describe the evolution of average activities of excitatory and inhibitory populations of neurons. We analyze the dependence of the model's behavior on two parameters. The parameter plane is partitioned into regions of equivalent behavior bounded by bifurcation curves, and the representative phase diagram is constructed for each region. This allows us to describe qualitatively the behavior of the model in each region and to predict changes in the model dynamics as parameters are varied. In particular, we show that for some parameter values the system can exhibit long-period oscillations. A new type of dynamical behavior is also found when the system settles down either to a stationary state or to a limit cycle depending on the initial point.

Biological Clocks↗

PubMed1992

Artificial neural network classification of Drosophila courtship song mutants.

Courtship songs produced by Drosophila males--wild-type, plus the cacophony and dissonance behavioral mutants--were examined with the aid of newly developed strategies for adaptive acoustic analysis and classification. This system used several techniques involving artificial neural networks (a.k.a. parallel distributed processing), including learned vector quantization of signals and non-linear adaption (back-propagation) of data analysis. "Pulse" song from several individual wild-type and mutant males were first vector-quantized according to their frequency spectra. The accumulated quantized data of this kind, for a given song, were then used to "teach" or adapt a multiple-layered feedforward artificial neural network, which classified that song according to its original genotype. Results are presented on the performance of the final adapted system when faced with novel test data and on acoustic features the system decides upon for predicting the song-mutant genotype in question. The potential applications and extensions of this new system are discussed, including how it could be used to screen for courtship mutants, search novel behavior patterns or cause-and-effect relationships associated with reproduction, compress these kinds of data for digital storage, and analyze Drosophila behavior beyond the case of courtship song.

Algorithms↗

PubMed1992

Universality in neural networks: the importance of the 'mean firing rate'.

We present a general analysis of highly connected recurrent neural networks which are able to learn and retrieve a finite number of static patterns. The arguments are based on spike trains and their interval distribution and require no specific model of a neuron. In particular, they apply to formal two-state neurons as well as to more refined models like the integrate-and-fire neuron or the Hodgkin-Huxley equations. We show that the mean firing rate defined as the inverse of the mean interval length is the only relevant parameter (apart from the synaptic weights) that determines the existence of retrieval solutions with a large overlap with one of the learnt patterns. The statistics of the spiking noise (Gaussian, Poisson or other) and hence the shape of the interval distribution does not matter. Thus our unifying approach explains why, and when, all the different associative networks which treat static patterns yield basically the same results, i.e., belong to the same universality class.

Mathematics↗

PubMed1992

Sensory segmentation with coupled neural oscillators.

We present a model of sensory segmentation that is based on the generation and processing of temporal tags in the form of oscillations, as suggested by the Dynamic Link Architecture. The model forms the basis for a natural solution to the sensory segmentation problem. It can deal with multiple segments, can integrate different cues and has the potential for processing hierarchical structures. Temporally tagged segments can easily be utilized in neural systems and form a natural basis for object recognition and learning. The model consists of a "cortical" circuit, an array of units that act as local feature detectors. Units are formulated as neural oscillators. Knowledge relevant to segmentation is encoded by connections. In accord with simple Gestalt laws, our concrete model has intracolumnar connections, between all units with overlapping receptive fields, and intercolumnar connections, between units responding to the same quality in different positions. An inhibitory connection system prevents total correlation and controls the grain of the segmentation. In simulations with synthetic input data we show the performance of the circuit, which produces signal correlation within segments and anticorrelation between segments.

Animals↗

PubMed1992

The emergence of long-lasting transients of activity in simple neural networks.

The question was investigated whether long-lasting transients of activity, observed to occur in the intact cerebral cortex (EEG slow (delta) waves and 'K' complexes) as well as in isolated tissues cultured in vitro, can also emerge in a model network of excitatory and inhibitory cells. We show that such transients can indeed occur even if the cells do not have built-in slow kinetics. For certain parameter settings, the network is in a bistable state in which periods of increased activity (long-lasting transients) alternate with minimal activity. Transients are triggered by spontaneously firing cells ('noise'), which, rather than via a build-up of recurrent synaptic inhibition, also initiate their termination. During a transient, the network continually makes transitions from one equilibrium to another as a result of spontaneous firing until it is switched back to the quiescent state, i.e., after a variable period of time of noise-induced transitions the transient is terminated. If the network is small, activity can terminate even without inhibition. In large networks, inhibition keeps the network sensitive to spontaneously firing cells by holding it in the neighbourhood of a critical point between active and quiescent state.

Action Potentials↗