Computational Neural Model Research Articles

Evolutionary origins and principles of distributed neural computation for state estimation and movement control in vertebrates

AbstractVertebrates are mostly large, agile creatures that can move rapidly and efficiently despite the significant inertial and interaction forces that are generated when they move. Vertebrates appear in the fossil record in the lower Cambrian, when predation began to apply selection pressure for increased size and, consequently, increased inertia. Agility depends critically upon the cerebellum, a brain structure that appears to have evolved as an elaboration of the vestibular and lateral‐line sensory processing regions in brains of aquatic ancestors. There is compelling evidence to suggest that the specific role of the cerebellum in motor control (etc) is state estimation. Analysis of responses of vestibular neurons in bullfrogs shows that these neurons have fractional‐order dynamics. Because of this property, the vestibular nerve forms a map of the state space of the animal's head, and individual action potentials can be regarded as assertions about the location of the head in that space. This view of vestibular coding suggests an analogy with particle filters, a new simulation‐based method for state estimation. In a particle filter the distribution of possible states consistent with observations is represented by a cloud of “particles” in a map of the state space. By regarding action potentials as particles in neural state estimators, we may develop a realistic model of neural computation in the cerebellum. At the same time—and independently of the biological verisimilitude of such a cerebellar model—we may develop useful new algorithms for state estimation. © 2005 Wiley Periodicals, Inc. Complexity 10: 56–65, 2005

Integrating neuroscientific data across spatiotemporal scales

A major challenge confronting neuroscientists is associated with the multiple spatial and temporal scales of investigation of neural structure and function. I shall discuss the use of computational neural modeling as one method to bridge some of the different spatial and temporal levels. This approach will be illustrated using large-scale, neurobiologically realistic network models of auditory and visual pattern recognition that relate neuronal dynamics to fMRI data. It will be demonstrated that the models are capable of exhibiting the salient features of both electrophysiological neuronal activities and fMRI values that are in agreement with empirically observed data. To cite this article: B. Horwitz, C. R. Biologies 328 (2005).

Acquiring and validating motion qualities from live limb gestures

This paper presents a neural computing model that can automatically extract motion qualities from live performance. The motion qualities are in terms of laban movement analysis (LMA) Effort factors. The model inputs both 3D motion capture and 2D video projections. The output is a classification of motion qualities that are detected in the input. The neural nets are trained with professional LMA notators to ensure valid analysis and have achieved an accuracy of about 90% in motion quality recognition. The combination of this system with the EMOTE motion synthesis system provides a capability for automating both observation and analysis processes, to produce natural gestures for embodied communicative agents.

A neurocomputational model for prostate carcinoma detection.

Current guidelines for prostate carcinoma screening rely primarily on the digital rectal examination (DRE) and prostate specific antigen (PSA). Well described patient risk factors for prostate carcinoma also include age, ethnicity, family history, and complexed PSA. However, due to the nonlinear relation of each of these variables with prostate carcinoma, it is difficult to predict reliably each patient's risk based on linear univariate analysis. The authors investigated a neural network to model the risk of prostate carcinoma by seven readily available clinical features. The database for the current study comprised 3268 men recently evaluated for the early detection of prostate carcinoma. The seven clinical features evaluated included age, race, family history, International Prostate Symptom Score (IPSS), DRE, and total and complexed PSA. Three hundred forty-eight subjects in the dataset included men with determined prostate biopsy outcomes and for whom at least 6 of 7 features were available. The dataset was divided randomly into a training set (60%) and a test set (40%), with n1/n2 cross-validation used to evaluate model accuracy, and was modeled with linear and quadratic discriminant function analysis and a neural computational system. After a model with acceptable goodness of fit was achieved, reverse regression analysis using Wilks's generalized likelihood ratio test was performed to evaluate the statistical significance of each input variable. The receiving operating characteristic (ROC) area for the neural computational system in the test set was 0.825, whereas total PSA and complexed PSA alone had ROC areas of 0.678 and 0.697, respectively. The ROC area of logistic regression in the test set was 0.510, linear discriminant function analysis was 0.674, and quadratic discriminant function analysis was 0.011. All were significantly less than the ROC area of the neural computational model (all Ps < 0.002). Reverse regression based on Wilks's generalized likelihood ratio test demonstrated each input feature to be highly significant to the model (all Ps << 0.000001). The authors modeled a combination of well described patient risk factors for prostate carcinoma using a neural computational system with acceptable goodness of fit. They demonstrated that each of the seven variates on which the model was based was critically significant to model performance. The authors presented this model for clinical use and suggested that clinicians use it in deciding to perform prostate biopsy.

Random Graphs In A Neural Computation Model

We examine in this work the following graph theory problem that arises in neural computations that involve the learning of boolean expressions by studying the asymptotic connectivity properties of $G_{n\\comma 1/\\lpar kn\\rpar ^{1/2}}$ random graphs, where k is a fixed positive integer. For an undirected graph $G = \\lpar V\\comma \\; E\\rpar $ let $N\\lpar X\\comma \\; Y\\rpar = \\lcub v \\in V - \\lpar X \\cup Y\\rpar \\!\\mid$ $ \\exists x \\in X\\ \\hbox{with}\\ \\lpar v\\comma \\; x\\rpar \\in E\\rcub $ . For fixed k construct an undirected graph $G = \\lpar V\\comma \\; E\\rpar $ such that for all disjoint sets $A\\comma \\; B \\subseteq V$ such that $\\vert A \\vert = \\vert B \\vert = k$ , and $C = N\\lpar A\\comma \\; B\\rpar \\cap N\\lpar B\\comma \\; A\\rpar $ , set C is such that $\\vert C \\vert$ is either exactly k or as close to k as possible. Asymptotic results for large values of k are also presented.

MSTd neuronal basis functions for the population encoding of heading direction.

Basis functions have been extensively used in models of neural computation because they can be combined linearly to approximate any nonlinear functions of the encoded variables. We investigated whether dorsal medial superior temporal (MSTd) area neurons use basis functions to simultaneously encode heading direction, eye position, and the velocity of ocular pursuit. Using optimal linear estimators, we first show that the head-centered and eye-centered position of a focus of expansion (FOE) in optic flow, pursuit direction, and eye position can all be estimated from the single-trial responses of 144 MSTd neurons with an average accuracy of 2-3 degrees, a value consistent with the discrimination thresholds measured in humans and monkeys. We then examined the format of the neural code for the head-centered position of the FOE, eye position, and pursuit direction. The basis function hypothesis predicts that a large majority of cells in MSTd should encode two or more signals simultaneously and combine these signals nonlinearly. Our analysis shows that 95% of the neurons encode two or more signals, whereas 76% code all three signals. Of the 95% of cells encoding two or more signals, 90% show nonlinear interactions between the encoded variables. These findings support the notion that MSTd may use basis functions to represent the FOE in optic flow, eye position, and pursuit.

Signal Processing Systems for Echolocation in the Bat's Brain

The neural base of the bat's biosonar system is discussed here. The bat emits complex biosonar sounds (pulses) and listens to echoes for orientation and hunting flying insects. Different types of biosonar information are carried by different parameters characterizing pulse-echo pairs. For example, distance information is conveyed by echo delay, while velocity information is carried by Doppler shift. In the auditory cortex of the bat, not only frequency but also other information baring parameters such as echo delay and Doppler shift are systematically mapped as subareas. These computational maps greatly depend on subcortical signal processing. The subcortical auditory nuclei create delay lines and multipliers (or AND gates) for processing distance (echo delay) information, and also create level-tolerant frequency tuning and multipliers (or AND gates) for processing velocity (Doppler shift) information. These multipliers are called FM-FM and CF/CF combination sensitive neurons, respectively. The neurophysiological investigations of the bat's biosonar system provide an excellent database for neural computational models and sonar systems. Application of an agonist of inhibitory neurotransmitter to DSCF or FM-FM area behaviorally revealed the functions of these auditory subareas. Findings made by an on-board telemetry microphone from flying bats confirmed Doppler-shift compensation.

A computational approach to control in complex cognition

Cognitive deficits associated with dorsolateral prefrontal cortex (DLPFC) damage are often most apparent in higher cognitive tasks that involve problem solving and managing multiple goals. However, computational models of prefrontal deficits on such tasks are difficult to construct. Problem solving is most naturally modeled with symbolic systems (e.g. production systems), but the effects of lesions are most naturally modeled with subsymbolic systems (neural networks). We show that when we adopt a simple and plausible model of neural computation, there is a natural and explicit mapping from symbolic, goal-driven cognition onto neural computation. We exploit this mapping to construct a neural network model that is capable of solving complex problems in the Tower of London task. The model leads to a specific hypothesis about the role of DLPFC in such tasks, namely, that DLPFC represents internally generated subgoals that modulate competition among posterior representations. When intact, the model accurately simulates the behavior of college students even on the most difficult problems. Furthermore, when the subgoal component is lesioned, it accurately simulates the behavior of prefrontal patients, including the fact that their deficits are most apparent on the most difficult tasks and that they have special difficulty with tasks that require inhibiting a prepotent response.

A neural computational model of ICSI outcomes from sperm source and other features

Objective: With the success and increasing popularity of ICSI, accurately predicting outcomes from this artificial reproductive technique becomes highly desirable. We examined a large dataset of ICSI outcomes and associated clinical features such as sperm source, which technician performed the procedure, number of ova retrieved and maternal age, and modeled outcomes using neural computational techniques. Neural computation employs computer programs that implement mathematical models based loosely on the function of networks of biological neurons. Statistical feature extraction, allowing examination of the significance of individual features such as sperm source, was employed using Wilk’s Generalized Likelihood Ratio Test (GLRT). Design: Neural computational modeling of retrospective data. Materials/Methods: We modeled fertilization outcomes from ICSI from technician (7 different technicians performed the procedure,) sperm source (testis extraction, MESA, EEJ,) number of ova retrieved and maternal age using neUROn2++, a set of C++ programs we developed to implement neurocomputational algorithms, discriminant function and regression techniques using the Cygwin (Red Hat) GNU C++ port for Windows (Microsoft) and Visual C++ (Microsoft) distributed across Pentium (Intel) platforms. Wilk’s GLRT was implemented in neUROn2++ for statistical feature extraction. A dataset of 1070 unique ICSI outcomes was randomized into a training set of 1349 exemplars and 449 test exemplars. N1/N2 cross-validation was employed. Results: Overall model accuracy was ROC AUC 0.922 (sensitivity 86.9%, specificity 79.0%) in the training set, and 0.897 (sensitivity 82.8%, specificity 75.1%) in the test set. Feature extraction using Wilk’s GLRT in reverse regression is shown in the table. All features were significant to the model (all p <1e-5), however, with p <1e-106, technician was highly significant.TableRegression analysis of neural computational model to predict ICSI outcomes.Featurep-valueTechnician1.04e-107No. ova retrieved9.84e-45Testis extraction3.81e-21Maternal age7.70e-20MESA2.33e-10EEJ1.52e-06p-values indicate significance of the individual feature to the likelihood of fertilization with ICSI. Open table in a new tab p-values indicate significance of the individual feature to the likelihood of fertilization with ICSI. Conclusions: We modeled ICSI outcome with high accuracy (ROC AUC 0.897 in the test set) using a neural computational approach. In order to determine the significance of individual features to the model’s outcome, regression based on Wilk’s GLRT was performed. All features examined (sperm source, number of ova retrieved, maternal age and which technician performed the procedure) were statistically significant to the model’s outcome. Supported by: NIH P01 HD36289 to Dolores J Lamb, Larry I Lipshultz and Craig Niederberger.

Texture segmentation performance related to cortical geometry

There are two prevailing explanations for the foveal deficit in texture segmentation reported in previous works. One is based on the spatial and temporal properties of the stimuli, which means in terms of physiology a strong contribution of the Magno-channel. The other one is purely spatial and assigns filters of different bandwidths to each eccentricity in the visual field. We have challenged the first explanation experimentally by using isoluminant stimuli. The central performance drop persisted although the Magno-channel is known to respond weakly to stimuli with low luminance contrast. Therefore, we agreed with the spatial explanation. But instead of the abstract filter theories from previous works we propose a computational neural model assuming local lateral interactions in a cortical map model. The psychophysical performance measures could be directly related to geometric properties of the primary visual cortex concerning its mapping geometry and its intrinsic interaction width. Our model accounts quantitatively for our own psychophysical data as well as for others from literature. In general, we claim that the high foveal retino–cortical magnification maps texture elements too far away from each other for being compared by local processes.

Energy-efficient neuronal computation via quantal synaptic failures.

Organisms evolve as compromises, and many of these compromises can be expressed in terms of energy efficiency. For example, a compromise between rate of information processing and the energy consumed might explain certain neurophysiological and neuroanatomical observations (e.g., average firing frequency and number of neurons). Using this perspective reveals that the randomness injected into neural processing by the statistical uncertainty of synaptic transmission optimizes one kind of information processing relative to energy use. A critical hypothesis and insight is that neuronal information processing is appropriately measured, first, by considering dendrosomatic summation as a Shannon-type channel (1948) and, second, by considering such uncertain synaptic transmission as part of the dendrosomatic computation rather than as part of axonal information transmission. Using such a model of neural computation and matching the information gathered by dendritic summation to the axonal information transmitted, H(p*), conditions are defined that guarantee synaptic failures can improve the energetic efficiency of neurons. Further development provides a general expression relating optimal failure rate, f, to average firing rate, p*, and is consistent with physiologically observed values. The expression providing this relationship, f approximately 4(-H(p*)), generalizes across activity levels and is independent of the number of inputs to a neuron.

Energy-Efficient Neuronal Computation via Quantal Synaptic Failures

Organisms evolve as compromises, and many of these compromises can be expressed in terms of energy efficiency. For example, a compromise between rate of information processing and the energy consumed might explain certain neurophysiological and neuroanatomical observations (e.g., average firing frequency and number of neurons). Using this perspective reveals that the randomness injected into neural processing by the statistical uncertainty of synaptic transmission optimizes one kind of information processing relative to energy use. A critical hypothesis and insight is that neuronal information processing is appropriately measured, first, by considering dendrosomatic summation as a Shannon-type channel (1948) and, second, by considering such uncertain synaptic transmission as part of the dendrosomatic computation rather than as part of axonal information transmission. Using such a model of neural computation and matching the information gathered by dendritic summation to the axonal information transmitted, H(p*), conditions are defined that guarantee synaptic failures can improve the energetic efficiency of neurons. Further development provides a general expression relating optimal failure rate, f, to average firing rate, p*, and is consistent with physiologically observed values. The expression providing this relationship, f approximately 4(-H(p*)), generalizes across activity levels and is independent of the number of inputs to a neuron.

A NEURAL COMPUTATION MODEL FOR REAL-TIME COLLISION-FREE ROBOT NAVIGATION

A biologically inspired neural computation model is proposed for dynamic planning and tracking control of robots. The dynamic environment is represented by a neural activity landscape of a topologically organized neural network, where each neuron is characterized by a shunting equation. The collision-free path is generated in real-time through the activity landscape without any prior knowledge of the dynamic environment. The real-time tracking control of robots to follow the planned path is also designed using shunting equations. The effectiveness is demonstrated through case studies. Simulation in several computer-synthesized virtual environments further demonstrates the advantages of the proposed approach.

Seeking an Integrated Model of Anesthetic Action

Seeking an Integrated Model of Anesthetic Action

Neural computation in urology: an orientation.

Neural computation is a field in which mathematical models are derived from algorithms based loosely on the physiological function of the biological neuron. This paper serves as an introduction to those that follow in this issue of Molecular Urology, which describe actual applications of neural computational modeling in the urologic domain. In this introductory paper, the history of computer technology and the foundations of neural computation are discussed. Methods of determining the accuracy of computation models are reviewed, and a statistical method of evaluating the significance of individual input features to the model's output, a process known as "feature extraction," is presented. Resources that provide free and commercial neural computational programs are cited for those readers interested in applying this technology to their own datasets, and a brief description of the author's neural computational programming environment is included. Finally, deployment of computational models via the Internet and various computer platforms is discussed.

Neural representation of alpha-oriented moving light bars in the cortex: a neural network study.

A neural computational model is suggested in this paper for investigating the stimulus dependence of spiking patterns and the neural representation of alpha-oriented moving light bars in the cortex. In this model, a stimulus-directed cortical developing algorithm is introduced for training the neural network. Three classes of computer simulations concerned with the orientation of the stimulus are carried out. The simulation results show that the fine temporal structure of spiking patterns of single units depends on the alpha orientation of the two moving light bars, and the fine temporal structure of their combinatorial spiking patterns are also context dependent. They also show that the neural representation of an alpha-oriented moving light bar is determined not only by the stimulus itself but also the architecture of the matured network. In the end, we propose a possible neural coding mechanism underlying the temporal cell subassemblies in the cortex, which could be spontaneously and dynamically organized into a dynamical cell assembly by synchronized activity of these subassemblies.

Prognosis of remaining bearing life using neural networks

A new concept referred to as progression-based prediction of remaining life (PPRL) is proposed in the present paper in order to solve the problem of accurately predicting the remaining bearing life. The basic concept behind PPRL is to apply different prediction methods to different bearing running stages. A new prediction procedure which predicts precisely the remaining bearing life is developed on the basis of variables characterizing the state of a deterioration mechanism which are determined from on-line measurements and the application of PPRL via a compound model of neural computation. The procedure consists of on-line modelling of the bearing running state via neural networks and logic rules and not only can solve the boundary problem of remaining life but also can automatically adapt to changes in environmental factors. In addition, multi-step prediction is possible. The proposed technique enhances the traditional prediction methods of remaining bearing life.

Modularity in neural computing

This paper considers neural computing models for information processing in terms of collections of subnetwork modules. Two approaches to generating such networks are studied. The first approach includes networks with functionally independent subnetworks, where each subnetwork is designed to have specific functions, communication, and adaptation characteristics. The second approach is based on algorithms that can actually generate network and subnetwork topologies, connections, and weights to satisfy specific constraints. Associated algorithms to attain these goals include evolutionary computation and self-organizing maps. We argue that this modular approach to neural computing is more in line with the neurophysiology of the vertebrate cerebral cortex, particularly with respect to sensation and perception. We also argue that this approach has the potential to aid in solutions to large-scale network computational problems - an identified weakness of simply defined artificial neural networks.

A simple model for neural computation with firing rates and firing correlations

AbsteractA simple extension of standard neural network models is introduced which provides a model for neural computations that involve both firing rates and firing correlations. Such an extension appears to be useful since it has been shown that firing correlations play a significant computational role in many biological neural systems. Standard neural network models are only suitable for describing neural computations in terms of firing rates.The resulting extended neural network models are still relatively simple, so that their computational power can be analysed theoretically. We prove rigorous separation results, which show that the use of firing correlations in addition to firing rates can drastically increase the computational power of a neural network.Furthermore, one of our separation results also throws new light on a question that involves just standard neural network models: we prove that the gap between the computational power of high-order and first-order neural nets is substantially larger than shown previously.

A graph-theoretic result for a model of neural computation

We deal in this work with the following graph construction problem that arises in a model of neural computation introduced by L.G. Valiant. For an undirected graph G = ( V, E), let set N∗(X, Y), where X, Y ⊂- V, denote the set of vertices other than those of X, Y which are adjacent to at least one vertex in X and at least one vertex in Y. An undirected graph G needs to be constructed that exhibits the following connectivity property. For any constant k and all disjoint sets A, B ⊂- V such that ¦A¦ = ¦B¦ = k, it is required that the cardinality of set C = N∗(A, B) be K or as close to k as possible. We prove that for k > 1, if any graph G exists so that for all choices of A and B, set C = N∗(A, B) has cardinality exactly k, then G must have exactly 3 k vertices. Thus an exact solution for arbitrarily large values of n does not exist for any such k. A graph construction based on a projective plane graph provides an approximate solution, in a certain sense, for arbitrarily large n.