CNS*1992

The Annual Computational Neuroscience Meeting

July 1992, San Francisco, California

CNS*1992 Abstracts

• Author: Gröbler, T.
  Biophysics Group,
  Central Research Institute for Physics,
  Hungarian Academy of Sciences, H-1525
  Budapest P.O. Box 49, Hungary

  Title: Dynamic phenomenia in the olfactory bulb II

  Abstract: A mathematical model has been given to demonstrate the associative memory character of the olfactory bulb. Odor qualities are coded in distributed spatial amplitude patterns. Differential equations for the mitral and granule cell activities have been supplemented by a continuous-time local learning rule. A nonlinear forgetting term and a selective decreasing term is added to the Hebbian learning rule. A learned odor can be recalled by a subset of the pattern. There is a strict restriction on the parameters: only those values can be admitted which generate physiologically justified activity signals.

  ---

• Author: Norberto M. Grzywacz,
  Smith--Kettlewell Eye Research Institute,
  2232 Webster Street, San Francisco, CA 94115
  Franklin R. Amthor,
  Department of Psychology and Neurobiology
  Research Center, University of Alabama at Birmingham,
  Birmingham, AL 3529

  Title: Gated--Enhancer Model for Motion Facilitation in Retinal Directional Selectivity

  Abstract: Directionally selective ganglion cells respond to preferred--direction motions more strongly than predicted by the sum of the responses to stationary stimuli covering the same area as the motion. The spatial, temporal, and contrast properties of this motion facilitation were recently investigated in rabbit by apparent--motion protocols. These properties are inconsistent with facilitation models based either on threshold or membrane--conductance modulation. However, these properties are consistent with models in which an enhancer agent induced by motion facilitates responses evoked later along the motion trajectory. The data indicate that the enhancer would itself not produce excitation until gated by the later excitation. Simulations using Surf--Hippo, a new retinal simulator, support a dendritic implementation of this ``gated--enhancer" model.

  ---

• Author: Michael E. Hasselmo, Michael Vanier Ross Bergman and James M. Bower
  M.E.H., R.B.
  Dept. of Psychology, Harvard Universtiy
  33 Kirkland St., Cambridge, MA 02138
  (617) 495-3875
  M.V., J.M.B.
  Div. of Biology 216-76, Caltech
  Pasadena, CA 91125

  Title: Cholinergic modulation of assoclative memory function in a realistic computational model of piriform cortex

  Abstract: A detailed model of piriform cortex developed with the GENESIS simulation package shows associative memory properties such as completion. This model allows analysis of the role of acetylcholine in cortical associative memory function. Experiments demonstrate that acetylcholine causes selective suppression of synaptic transmission at excitatory intrinsic fiber synapses in piriform cortex (Hasselmo and Bower, 1992). When applied during learning in the model, this selective cholinergic suppression prevents interference between overlapping input patterns, allowing separate and distinct learning of input stimuli. This selective suppression may act in concert with the cholinergic increase in excitability to modulate piriform cortex associative memory function.

  ---

• Author: P.G.Hearne, J.Wray, E.Agar, D.J.Sanders and G.G.R.Green
  Department of Physiological Sciences,
  The Medical School,
  Newcastle upon Tyne, NE2 4HH. UK

  Title: The Neurone Describing function : A Single Compartment Study

  Abstract: In artificial neural networks, units usually perform a sigmoid function of the weighted sum of their inputs. By Taylor series expansion, it can be shown that these networks compute a constrained Volterra series of their inputs. As a descriptor of real neurone behaviour this approach fails because neurones have dynamic properties and do not have a separate linear stage. Using compartmental modelling, describing functions of neurones can be approximated. This approach demonstrates a frequency dependent non-linearity not represented by a fixed Volterra Kernel. The simulation of real neurones leads to the proposal of alternative describing functions for artificial nets.

  ---

• Author: Michael Herrmann
  Dept. of Computer Science,
  Leipzig University, 0-7010 Leipzig, F.R.G.
  Eytan Ruppin
  Department of Computer Science,
  Raymond and Beverly Sac~;ler Faculty of Exact Sciences,
  Tel Aviv University, Tel Aviv 6997O, Israel
  Marius Usher
  CNS program, Div. of Biology 216-76,
  Caltech, Pasadena, CA91125

  Title: A network for semantic and episodic associations revealing thought disturbances due to neural loss

  Abstract: We study an Attractor Neural Network that stores natural concepts, organized in semantic classes. The concepts are represented by distributed patterns over a space of attributes, and are related by both semantic and episodic associations. While semantic relations are expressed through an hierarchical coding over the attribute space, episodic links are realized via specific synaptic projections. Due to dynamic thresholds expressing neuronal fatigue, the network's behavior is characterized by convergence toward the concept patterns on a short time scale, and by transitions between the various patterns on a longer time scale. The network manifests semantic, episodic, and random transitions. Modeling pathological memory disturbances, we studied the influence of several parameters on the frequency of semantic and episodic transitions. W'hen neurons characterized by a large synaptic connectivity are deleted, semantic transitions deteriorate before the episodic ones, in accordance with the findings in patients with Alzheimer's disease.

  ---

• Author: Marty Hiller
  MIT Artificial Intelligence Laboratory
  545 Technology Square, NE43-825
  Cambridge, MA 02139

  Title: A Model of the Combined Effects of Chemical and Activity-Dependent Mechanisms in Topographic Map Formation

  Abstract: A simulated model of the effects of chemical and activity-dependent mechanisms in to- pographic map formation is presented. These mechanisms work together to allow rapid development of detailed maps. The model matches experimental results on the ability to form rough, but not refined, maps in the absence of electrical activity, expansion of heavily stimulated areas, and formation of ocular dominance columns. This work explains and resolves problems with previous models, including the need to use non-biological input representations, slow convergence rates, and problems with orientation and formation of discontinous sub-maps.

  ---

• Author: D. Horn & E. Ruppin
  Raymond and Beverly Sackler Faculty of Exact Sciences,
  Tel Aviv University, Tel Aviv 69978, Israel
  M. Usher
  CNS program, Div. of Biology 216-76,
  Caltech, Pasadena, CA91125
  M. Herrmann
  Dept. of Computer Science,
  Leipzig University, 0-7010 Leipzig, F.R.G.

  Title: Synaptic deletion and compensation in Alzheimer disease: A neural model

  Abstract: Recent experimental evidence testifies to the possible role of synaptic deletion and compensation in the pathogenesis of Alzheimer's disease (AD). As a clinical hallmark of AD is memory degradation, these processes are examined in the framework of a neural network model of associative memory. Using a network with fixed memory load we vary two parameters, the deletion of synapses and the total synaptic strength, and observe the characteristic turnover from a memory-retrieval phase to total memory loss. Several synaptic compensation strategies, that patients may employ depending on their synaptic regenerative potential, are specified and examined. Reviewing the neuroanatomical and clinical data, we show that our model can account for the variation in the severity and progression rate of AD.

  ---

• Author: Gwen Jacobs, Frederic Theunissen, Jay Levin and Michael Landolfa,
  Department of Molecular and Cell Biology,
  University of California, Berkeley, CA 94720.

  Title: Morphological basis of wind stimulus discrimination in the cricket cercal sensory system

  Abstract: The functional organization of a sensory map in a simple insect sensory system was studied using a combination of anatomical and electrophysiological techniques. Analysis of the map indicates that air current direction is represented in a continuous fashion within the map. Directionally selective primary sensory interneurons extract directional information from the map by virtue of the position of their dendritic arbors within the map. To test this structure-function hypothesis we have modeled the directional tuning curves of primary sensory interneurons based on the anatomical relationships between the interneurons and the map. The predicted tuning curves are very similar to the experimentally observed curves.

  ---

• Author: Dieter Jaeger, Erik De Schutter and James M. Bower,
  California Institute of Technology,
  Div. of Biology 216-76, Pasadena CA 91125

  Title: Prolonged depolarizations following brief synaptic stimulations in the cerebellar Purkinje cell: Intracellular recordings and compartmental modelling

  Abstract: Intracellular recordings were obtained from Purkinje cells in the guinea pig slice preparation. Electrical stimulation pulses of 0.1 ms duration delivered to the granule cell layer resulted in 100 to 300 ms lasting depolarizations in Purkinje cells. During these depolarizations the cells often showed an increase of spiking frequency. Pulses of synchronous synaptic activation in a compartmental model of the Purkinje cell resulted in similar prolonged depolarizations. In the model, prolonged depolarizations are due to the opening of voltage dependent calcium channels. Increases of spiking frequency with a similar time course can be observed in vivo with brief (5ms) tactile stimulation, suggesting that the active membrane properties of individual Purkinje cells observed in vitro and in the model contribute to the modulatiGn of Purkinje cell activity via the parallel fiber system in the functioning network.

  ---

• Author: Mark James and Doan Hoang
  Basser Department of Computer Science, F09
  The University of Sydney NSW 2006
  AUSTRALIA

  Title: Outline of a Theory of Isocortex

  Abstract: We outline a theory of six-layered, homeotypic cortex that includes mechanisms for pattern selec- tivity, hierarchical category recognition and sequence learning. Learnable inhibitory feedback between layers of cortex acts to shift the activation threshold for pattern-recognizing cells, causing them to become tuned to specific patterns of afferents. The final amount of inhibition will be a function of the "size" of the pattern recognized by the cell. The inhibition can therefore be applied to other pattern-recognizing cells to implement a "masking field" (Grossberg, 1985). Grossberg, S., The adaptive self-organization of serial order in behavior: Speech, language, and motor control. In Schwab, E.C. and Nusbaum, H.C. (Eds.), Perception of speech and visual form: Theoretical issues, models, and research, Academic Press, NY, 1985.

  ---

• Author: Janani Janakiraman K.P.UnniKrishnan
  Advanced Technology Laboratories Computer Science Department
  The University of Michigan GM Research Laboratories
  1101 Beal Avenue, Ann Arbor, MI 48109 Warren, MI 48109-9055

  Title: A model for dynamical aspects of visual attention

  Abstract: A model is presented to investigate the role of feedback pathways from association cortical areas to Vl and LGN in attentional mechanisms. In the model,these pathways are shown to be capable of modifying the tuning properties of lower level neurons, hence quickening their convergence. Model simulations use an algorithm (Alopex) that is implementable by known neural circuitry. The change in tuning properties of V4 and IT neurons in the monkey has lead to the hypothesis that these changes play a role in the static aspects of attention. Our results show that this may instead be a dynamic phenomenon,helping the visual system to focus attention rapidly.

  ---

• Author: R.V. Jensen* and G.M. Shepherd
  Department of Neurobiology,
  Yale University School of Medicine New Haven,
  CT 06510 and Department of Physics,
  Wesleyan University Middletown, CT 06457

  Title: NMDA-Activated Conductances Provide Short-Term Memory for Dendritic Spine Logic Computations

  Abstract: Active conductances in dendritic spines may permit elaborate computational processing of multiple synaptic inputs long before these signals reach the soma. Numerical models of dendritic trees indicate that the interactions of postsynaptic potentials in active spines can generate simple logic operations such as AND, OR and NAND gates. However, because the spine head EPSP's closely follow the underlying, short-duration (1-3 ms), synaptic conductances, previous studies concluded that the precise timing of synaptic inputs is critical for these logic operations. We show that this severe temporal limitation on dendritic computation can be overcome by the inclusion of slow (100-300 ms), voltage-dependent, NMDA-receptor mediated conductances$^5$ in the spine heads. Our numerical simulations show that this simple mechanism provides a short term memory ($\sim$ 100 ms) for logical AND gates with time-delayed inputs on one or more spines.

  ---

• Author: R.V. Jensen, A. Williamson, D. Berkowicz, G.M. Shepherd
  Department of Neurobiology,
  Yale University School of Medicine
  New Haven, CT 06510

  Title: NMDA Receptors and Hyperexcitability in Human Hippocampal Granule Cells

  Abstract: Hippocampal granule cells from patients with certain types of temporal lobe epilepsy have been found to exhibit abnormal electrophysiological and pathological properties. In particular, intracellular recordings in slice preparations from human patients with medial temporal lobe sclerosis show an enhacement of the NMDA component of the EPSP relative to normal human or rodent tissues.1 In addition, the axons of the granule cells in the diseased tissue have been observed to sprout recurrent fibers back to the proximal dendritic layers of the granule cells.2 A comparison of the results of the experimental measurements with the analysis of numerical multi-compartmental models incorporating the electrophysiological properties of these abnormal cells indicates a possible mechanism by which the enhanced NMDA mediated conductance and the axonal feed-back can lead to the observed hyperexcitability of these cells that may be responsible for epileptic activity.

  ---

• Author: Yoshimi Kamiyama, Hiroyuki ISHII and Shiro USUI
  Department of Information and Computer Sciences
  Toyohashi University of Technology
  Toyohashi 441, JAPAN

  Title: A Method for Estimating the Neural Input to a Neuron using the Ionic Current Model

  Abstract: The ionic current properties of a neuron play a key role in information processing in neural systems. These ionic currents can be described mathematically similar to Hodgkin-Huxley equations based on voltage clamp experiments. Such an ionic current model provides a basis for understanding the neuronal function and for calculating the individual ionic currents from experimentally recorded voltage responses. Here, we propose a method to estimate the total input current to the neuron which produces the voltage response by using the model. We first applied the method to Hodgkin-Huxley model and confirmed that the input current is estimated correctly. Secondly, we applied the method to retinal horizontal cell response and analyzed the estimated total input current for the voltage responses with various light stimuli. Finally we discussed the role of individual ionic current in the horizontal cell membrane and the underlying mechanisms of the nonlinear dynamical voltage response.

  ---

• Author: R.W. Kentridge,
  Department of Psychology,
  University of Durham,
  Durham DH1 3LE,
  U.K.

  Title: Dissipative Structures and Self-Organizing Criticality in Neural Networks with Spatially Localized Connections.

  Abstract: Neural networks with asymmetric random connections may converge to simple attractors or they may behave chaotically. If, however, the probability of neu- ral interconnection in a network is related to the distance between neurons, as it is in the brain, then a third class of behavior (which could be used in coding representation), dissipative structure formation, becomes possible. Simulations of such networks showed that dissipative structures could form, however, this required biologically implausible fine-tuning of the networks' parameters. Further studies showed that conditions in which dissipative structure formation was likely could be produced by uniform low intensity stimulation of these networks through a process of self-organizing criticality. This mechanism eliminates the requirement for biologically implausible fine-tuning of networks thereby increasing the probability that dissipative structures may play a functional role in processing information in the brain.

  ---

• Author: Garrett T. Kenyon and David C. Tam
  Division of Neuroscience, Baylor College of Medicine,
  One Baylor Plaza, Houston, TX 77030

  Title: Deterministic Entropy Measures for Investigating Complex SpatioTemporal Order in Multi-Unit Spike Train Data

  Abstract: We consider the hypothesis that within a neural population, the distribution of cell firing times is chaotic, characterized by a finite Kolmogorov entropy. We contrast this with the alternative rate code hypothesis, which asserts that the distribution of cell firing times is ergodic, possessing an infinite Kolmogorov entropy. Starting from an explicit expression for the entropy of a train of uncorrelated events, characteristic features for the entropy of any ergodic process are inferred. Entropy growth that is slower than logarithmic at increased temporal resolutions, as well as anomalous renormalization flow, is argued to be a sufficient indicator of an underlying chaotic order. Stereotypical results are illustrated through the analysis of computer generated spike train data. A parallel algorithm for extracting the spike train entropy is provided.

  ---

• Author: Daryl R. Kipke, Erik Herzog, and Robert B. Barlow, Jr.
  Institute for Sensory Research, Syracuse University

  Title: A Computational Model of the Limulus Lateral Eye on a Parallel Computer

  Abstract: The lateral eye of Limulus provides an unique opportunity to investigate retinal function and the neural code of vision. We have integrated computational, physiological, and behavioral experimental methods to investigate the information the eye sends to the brain and the efferent modulation of the eye by the brain. We developed a cell-based computational model of the eye on a parallel computer (Connection Machine). The model structure is related closely to the anatomical structure of the eye and the parameters are estimated from experimental data. Model computations are combined with behavioral data to investigate the neural representation of the animal's visual world.

  ---

• Author: Peter Konig, Bernd Janosch, and Thomas B. Schillen
  Max-Planck-Institut fur Hirnforschung
  Deutschordenstrasse 46,
  6000 Frankfurt 71, FRG

  Title: Assembly Formation and Segregation by a Self-organizing Neuronal Oscillator Model

  Abstract: Experimental evidence demonstrates the stimulus-dependent formation and segregation of neuronal assemblies defined by coherent oscillatory response patterns. In this paper, we investigate whether the self-organization of synchronizing and desynchronizing connections can establish a corresponding temporal response structure using local learning rules. Motivated by recent experimental observations, synchronizing connections are modified according to a two-threshold Hebb-like learning rule, while we generalize this rule to analogous Anti-Hebb-like weight changes for the desynchronizing connections. We show that the resulting network exhibits synchronization and segregation of oscillatory activity in agreement with the experiment.

  ---

• Author: T. Kumar,
  Department of Molecular and Cell Biology*
  D. A. Glaser,
  Departments of Physics and of Molecular and Cell Biology
  337 Stanley Hall,
  University of California, Berkeley, CA 94720

  Title: Absence of Learning in Hyperacuity

  Abstract: Human observers with previous experience in psychophysical methods were tested on 34 different non-stereo and 33 stereo hyperacuity tasks. None of the observers had any previous experience with non-stereo hyperacuity task. The preliminary results showed no learning for the non-stereo tasks. The initial performancesQdefined by pooling the first five responses for each of the 34 stimuliQwere well within hyperacuity range. All observers had experience in stereo discrimination; their pooled initial five to ten responses indicated no initial practice effects for stereo discrimination in the fixation plane, but quite a rapid learning rate for stereo discrimination 5' off the fixation plane. Mechanisms of visual hyperacuity do not appear to require learning or other processes to improve performance with practice.

  ---

• Author: I Kupfermann, D. Deodhar, S.R. Rosen, and K.R. Weiss*.
  Center for Neurobiology and Behavior,
  Columbia University, 722 W. 168
  St. New York, NY 10032
  and * Fishberg Center, Mt. Sinai
  School of Medicine, 1 Gustave Levy Plaza,
  New York, NY 10029.

  Title: The use of genetic algorithms to explore neural mechanisms that optimize rhythmic behaviors: Quasi-realistic models of feeding behavior in Aplysia

  Abstract: Using simple neural models, we have been studying how behavior (feeding) can be optimized. The effects of circuit parameters are studied by exhaustive search or genetic algorithms. The findings indicate: 1) A defined two neuron-two muscle system can generate rhythmic behavior that can result in a net gain of energy. 2) In the absence of modulatory inputs, the efficiency of the system sharply degrades when individual parameters are varied only a small percentage from those used to evolve the system. It is postulated that various forms of circuit modulation will improve the functioning of the system over a wide range of external and internal parameters.

  ---

• Author: Gwendal Le Masson, Eve Marder and L.F. Abbott
  Department of Biology and Physics and Center for Complex Systems
  Brandeis University Waltham, MA 02254

  Title: A model for self-regulation of Neuronal activity

  Abstract: We constructed a conductance-based model from biophysical data of a stomatogastric ganglion neuron. Using this model, we address the question of stability of neuronal activity despite changes in both the internal and external environment. We propose that calcium could act as a regulatory signal to adjust the ratio of conductances expressed by a neuron. We show that the internal calcium concentration is highly correlated with electrical properties and we build a feedback mechanism to link calcium to the maximal conductance of each current. Models with dynamical maximal conductances can self-assemble the currents they need to produce a given target activity pattern from an arbitrary initial state and can recover from drastic changes in extracellular ionic concentration. This mechanism allows increased robustness and activity-dependent stabilization of network function.

  ---

• Author: Zhaoping Li
  School of Natural Science, Institute for Advanced Study
  Princeton, NJ 08540, USA

  Title: Understanding the Functional Goals of Retinal Ganglion Cells

  Abstract: In mammalian retina, the Y (or M) ganglion cells are significantly more transient in response, more selective to stimuli of low spatial and high temporal frequencies, and less selective to spectral information than the X (or P) cells. It is shown that these differences in cell properties can be explained by a model that assigns different functional goals to the different ganglion cell types. In this model, the goal of the Y cells is to extract as fast as possible the minimum amount of information necessary for quick responses. In contrast, the goal of the X cells is to extract as much information as possible. Temporal characteristics of the information extraction by the two cell groups are also derived.

  ---

• Author: Hans Liljenstrom and Michael E. Hasselmo
  SANSÑStudies of Artificial Neural Systems
  Dept. of Numerical Analysis and Computing Science
  Royal Institute of Technology, S-100 44 Stockholm, Sweden
  Dept. of Psychology
  Harvard University
  Cambridge, MA 02138, USA

  Title: Acetylcholine and cortical oscillatory dynamics

  Abstract: Acetylcholine appears to regulate the oscillatory properties of cortical structures. Application of cholinergic agonists induces theta rhythm oscilla- tory patterns in the piriform (olfactory) cortex (Biedenbach, 1966) and the hippocampus (Konopacki et al., 1987). These effects appear due to alter- ation of the intrinsic oscillatory characteristics of cortical structures. In a simulation of the oscillatory dynamics of piriform cortex (Liljenstrom, 1991), we show that the neuromodulatory effects of acetylcholine can induce theta rhythm oscillatory characteristics. The cholinergic increase in excitability causes increased oscillatory activity while the cholinergic suppression of ex- citatory intrinsic synaptic transmission prevents gamma rhythm oscillatory dynamics from predominating.

  ---

• Author: C. Linster*, C. Masson**, M. Kerszberg***, L. Personnaz*, G. Dreyfus*
  * Ecole Superieure de Physique et de Chimie Industrielles de la Ville de Paris,
  Laboratoirc d'Eleclronique
  10, rue Vauquelin
  75005 PARIS - FRANCE
  ** Laboratoire dc Neurobiologie Comparee des Invertebres
  INRA/CNRS
  91140 BURES SUR YVETTE - FRANCE
  ***Institut Pasteur
  25, rue du Docteur Roux
  75015 PARIS

  Title: Formal Model of the Insect Olfactory Macroglomerulus

  Abstract: An original model of the specialist olfactory system of insects, built in a semi-random fashion constrained by biological data is presented. A classification of the response patterns of individual neurons, based on the temporal aspects of the observed response, allows to analyze the behavior of the model with respect to he qualitative and the temporal variations of the stimulus. Among the results, a number are related to data about olfactory information processing reported in the literature, others may serve as guidelines for further investigations. We show how the model computes the information relevant as a basis for further processing in higher centers.

  ---

• Author: S. R. Lockery, S. J. Nowlan and T. J. Sejnowski
  Computational Neurobiology Laboratory
  The Salk Institute for Biological Studies
  P. O. Box 85800
  San Diego, CA 92186-5800
  and The Howard Hughes Medical Institute

  Title: Modeling Chemotaxis in Simple Nervous Systems

  Abstract: Parameters in network models can be adjusted using algorithms that optimize the fit between the response of neurons in the model and biological system. Often, however, more is known about behavior than neuronal responses. Using chemotaxis in the nematode C. elegans as a model system, we adapted a recurrent backpropagation algorithm to optimize parameters using behavioral data alone. In this form, backpropagation successfully optimized a dynamical network to chemotax in a simple environment. This sets the stage for the use of optimization techniques in more realistic models of the anatomically well-defined nervous system of C. elegans.

  ---

• Author: William W. Lytton Terrence J. Sejnowski
  Computational Neurobiology Laboratory
  The Salk Institute
  10010 N. Torrey Pines Rd.
  La Jolla, CA 92037

  Title: Intrinsic oscillations in simulated thalamocortical relay cells

  Abstract: Thalamocortical relay cells show a rich repertoire of oscillatory behaviors depending on membrane voltage: 1- 3 Hz slow wave oscillations, 8-10 Hz spindle oscillations and 100 Hz repetitive spiking. We have looked at the intrinsic dynamics of a model relay cell with realistic geometry and 9 voltage-sensitive channels. Spindling and slow wave oscillation could be reproduced in the model but were just two points along a spectrum of oscillation rate that varied with degree of hyperpolarization. Simulated cortical stimulation was effective ln phase resetting of either rhythm and could entrain the slower rhythm to higher frequencies.

  ---

• Author: Y. Matias
  Tel Aviv University & University of Maryland
  Address: UMIACS, University of Maryland,
  College Park, MD 20742, USA.
  E. Ruppin
  Department of Computer Science,
  Raymond and Beverly Sackler Faculty of Exact Sciences,
  Tel Aviv University, Tel Aviv 69978, Israel

  Title: A Neural Model for a Randomized Frequency-Spatial Transformation

  Abstract: This work addresses the question whether information encoded in the neuronal firing frequency can be transformed into spatial encoding. If neurons with a broad spectrum of membrane time constants (MTC) exist, or a precisely wired circuitry incorporating delay lines is assumed, a frequency-spatial transformation (FST) is clearly possible. However, various cortical regions are composed of neurons with short MTCs, which are randomly connected with each other. Introducing a stochastic component into the neurons' dynamics, we present a FST scheme which incorporates random connections and requires memory in a limited sense. Our scheme is presented and analyzed for a feed-forward network. We also discuss its applicability for a random network, and present some supporting simulation results.

  ---

• Author: Bartlett W. Mel
  Computation and Neural Systems Program
  Division of Biology, 216-76
  California Institute of Technology
  Pasadena, California 91125

  Title: Large Nonlinear Pattern Discrimination in Dendritic Trees

  Abstract: A recent compartmental modeling study showed that a dendritic tree rich in NMDA-type synaptic channels is selectively responsive to spatially clustered, as opposed to diffuse, pattens of synaptic activation \cite{mel:neuralcomp}. This ``cluster-sensitivity'' property was shown to be a sufficient mechanism to underlie true nonlinear pattern discrimination at the single cell level. The effects of other physiologically-characterized nonlinear membrane mechanisms have now been tested with respect to dendritic cluster-sensitivity, including voltage-dependent sodium and calcium channels of various kinetics and in various spatial distributions. Under loose assumptions, the capacity for nonlinear pattern discrimination can be greatly enhanced by the presence of excitatory voltage-dependent membrane channels. An abstract model neuron, called a {\sl clusteron}, is shown to capture important aspects of the cluster-sensitive input-output behavior of a full biophysically-modeled dendritic tree. An analysis of the storage capacity of the {\sl clusteron} is presented, and empirical results with a cluster-sensitive Hebb-type learning rule are shown to lead to a large storage capacity for nonlinearly-separable pattern sets.

  ---

• Author: Edmond Mesrobian and Josef Skrzypek
  Machine Perception Laboratory
  Computer Science Department
  University of California
  Los Angeles, CA 90024

  Title: Adaptive Receptive Fields for Textural Segmentation

  Abstract: Textural segmentation plays an important role in the figure-ground discrimination process. Many current models of mechanisms underlying textural segmentation assume that texture "contrast" information, captured at the border of two abutting textured regions, is sufficient to reconstruct the original surfaces of the textures. These approaches are based on the additional assumption that center-surround receptive fields of visual neurons are perfectly balanced; a uniform stimulus pattern, covering both the center and the surround, produces a zero response. We argue that such models would produce incorrect reconstructions of textural surfaces, and describe a neural network architecture for textural segmentation that can adaptively delimit the boundaries of uniformly textured regions. Simulation results are presented to demonstrate the segmentation capabilities of the architecture.

  ---

• Author: J.P. Miller, F. H. Eeckman and F.E. Theunissen.
  Dept of Mol. and Cell Biology.
  University of California, Berkeley. CA 94720

  Title: Linearization by Noise and Shunting Current of a Modified Hodgkin-Huxley Spiking Model.

  Abstract: We have developed a detailed description of voltage dependent ion channels to model the currents involved in spike generation in complex neuron models. The model consists of a reduced Hodgkin-Huxley formalism with added noise and shunting terms. Phase plane analysis was introduced to study the behavior of this model under various parameter settings. We used this model to reproduce the spike shape and spiking statistics of one particular type of interneuron in the cricket cercal sensory system, for which a large amount of electrophysiological and morphological data is available.

  ---

• Author: Read Montague Peter Dayan Terrence J. Sejnowski
  Computational Neurobiology Laboratory
  The Salk Institute
  P.O.Box 85800
  San Diego, CA 92186-5800

  Title: Signaling Local Synaptic Covariance through Space

  Abstract: Recent experimental evidence suggests that a rapidly diffusing signal produced at active synapses may be a determiner of Hebbian plasticity. We have analysed this new mechanism and have considered examples of the different predictions such a local signaling mechanism forces: (1) explicit normalisation is not required for the development of ocular dominance columns; (2) there is a critical minimum column width; (3) the diffusible signal produced at active synapses could be expected to modulate the efficacy of other synaptic terminals in the region containing different neurotransmitters; and (4) if the mechanism can act over short time-scales, then it can boost the signal/noise ratio through local contrast enhancement. We demonstrate these conclusions through simulations performed on an MIMD hypercube.

  ---

• Author: Venkatesh N. Murthy and Eberhard E. Fetz
  Dept. of Physiology & Biophysics,
  SJ-40, University of Washington,
  Seattle, Wa 98195

  Title: Effect of input synchrony on the response of a model neuron

  Abstract: We studied the dependence of the average spike frequency (f-out) of a biophysically realistic model neuron on the degree of synchrony (s, varied from 0 to 100%) in its excitatory synaptic inputs. The following independent parameters were systematically varied: (1) N, the number of inputs to the neuron, (2) f-in, the average frequency of the inputs, and (3) w, the strength of the input connection. The goal was to empirically determine the relation: f-out = F(N, fin, w, s). The results indicate that for parameter values in the range found in neocortical cells, input synchrony can affect output significantly.

  ---

• Author: Ernst Niebur
  Computation and Neural Systems Program
  California Institute of Technology
  Pasadena CA 91125, USA
  Florentin Worgotter
  Institut fur Physiologie, Ruhr-Universitat Bochum
  D-4630 Bochum, Germany

  Title: The architecture of visual cortical orientation columns

  Abstract: The dimension reduction from the multi-dimensional feature space to the two-dimensional cortical plane is realized by complex maps which so far have evaded intuitive understanding. We show that the most salient features of these maps can be understood from a few basic design principles: local correlation, isotropy and homogeneity. We define these principles in a mathematically exact sense and we show that they are sufficient to generate realistic column structures. We suggest that the reason why most of the previously proposed models of orientation columns are capable of generating realistically looking orientation column structures is simply that any model which is consistent with the aforementioned principles is capable of doing so.

  ---

• Author: Harmon S. Nine K.P. Unnikrishnan2
  Advanced Technologies Laboratories Computer Science Department
  The University of Michigan GM Research Laboratories
  1101 Beal Ave. 30500 Mound Rd.
  Ann Arbor, MI 48109 Warren, MI 48090-9055

  Title: The role of subplate feedback in the development ocular dominance columns

  Abstract: A model is presented to investigate the role of the feedback in the development of ocular dominance columns (ODC's) in V1. In the model, the the subplate locally integrates the activity of cortical layer IV is feeds it back to the same. Anatomical evidence indicates that this is possible. Modification of geniculocortical synaptic strengths is accomplished by the Alopex algorithm, which utilizes temporal correlations between changes in the feedback activity and changes in geniculocortical input activity. Alopex could be implemented by geniculocortical NMDA receptors in conjunction with subplate feedback. Computer simulations of the model resulted in robust formation of ODC's.

  ---

• Author: Michael I. Miller
  The Institute for Biomedical Computing and
  The Department of Electrical Engineering
  Washington University at St. Louis
  Campus Box 1161
  St. Louis, Missouri 63130
  Andrew T. Ogielski
  Bell Communication Research
  445 South St.
  Morristown, New Jersey 07960

  Title: A model for synaptic transmission in the auditory nerve: randomness, plasticity and signal estimation

  Abstract: In the cochlea of the cat and other mammals each afferent auditory neuron receives all its input from a single synapse on its target inner hair cell. The randomness and plasticity of synaptic transmission therefore profoundly influence the encoding of sound by action potentials in the auditory nerve. We present a mathematical model of continuous signal transmission by a synapse that can account for the observed history dependence and statistics of auditory nerve firing. The operation of a synapse is described in terms of interacting stochastic processes - stimulus driven vesicle exocytosis, vesicle recirculation, and postsynaptic potential generation.