CNS*1994

The Annual Computational Neuroscience Meeting

July 1994, Monterey, California

CNS*1994 Abstracts

Author: Leslie C. Osborne

Group in Biophysics
University of California
Berkeley, CA 94720

Title: MEASUREMENTS OF CRICKET WIND-SENSITIVE HAIR MOTION WITH LASER FEEDBACK INTERFEROMETRY

Abstract: The wind-sensing hairs of crickets are an accessible mechanosensory system in which one can quantify both the stimulus and the response and thereby extract the mechanical filter that passes air-motion inforrnation to the neurons of the cercal system. Tuning and range fractionation are accomplished by size variation in the hairs and the fluid mechanical properties of the cercus. Using laser feedback interferometry, I measured hair motion down to the level of Brownian motion which may represent a fundamental noise source in the neural computations of the cercal system.
Author: Bruce R. Parnas* and Muriel D. Ross

Biocomputation Center
MS 239-11, NASA Ames Research Center
Moffett Field, CA 94035-1000

Title: A 3-D INTERACTIVE MODEL FOR PERIPHERAL VESTIBULAL SIGNAL PROCESSING

Abstract: We have developed a 3-D interactive abstract network model for the vestibular periphery. This model allows the user to rotate, translate and scale the model so that specific portions may be viewed. This feature becomes increasingly important as the size of the system being simulated increases. In addition the user may choose specific component elements and see the associated temporal waveforms and receptive fields in separate windows. The model uses simple representations for anatomical elements, resulting in a computationally efficient system. With this system we hope to model the effects of altered gravity experiments on signal processing in the vestibular periphery.
Author: U. Ernst, K. Pawelzik*, and T. Geisel

Institut fur Theoretisch Physik, Universitat Frankfurt
60054 Frankfurt/M., Germany

Title: MULTIPLE PHASE CLUSTERING OF GLOBALLY PULSE COUPLED NEURONS WITH DELAY

Abstract: Neuronal synchronizations have gained increased attention since it has been suspected that they related to higher brain functions. The basic mechanisms leading to synchronization and desynchronization in realistic neuronal groups are still only insufficiently understood. In this contribution we analyse the mathematical gronds for synchronization in groups of simple pulse coupled neurons with finite transmission delays. Finite transmission delays and pulselike coupling yield a rich phenomenology including multiple clustering and spontanous desynchronization. In particular we analyse the stability of synchronization and derive a simple mechanism leading to synchronization in case of inhibition. Our treatment explains the emergence of multiple phase clustering for delayed mutual inhibition while for delayed excitatory interaction we can prove that precise synchronization is unstable.
Author: S. F. Peyton* 1 and D. R. Kipke 2

Arizona State University
Bioengineering Program
Tempe, AZ 85287-6006

Title: A COMPARTMENTAL MODEL OF VENTRAL COCHLEAR NUCLEUS STELLATE CELLS: RESPONSES TO CONSTANT AND AMPLITUDE-MODULATED TONES

Abstract: ln this study, we developed biologically plausible, compartmental models of ventral cochlear nucleus stellate cells to investigate the effects of synaptic distribution on spike-discharge activity using constant and amplitude-modulated tones. Differences in the spatial distribution of excitatory auditory nerve (AN) inputs and inhibitory non-cochlear inputs evoke a range of post-stimulus time histograms (PSTHs) chopper responses from sustained to transient to onset. These simulations illustrate how non-linear dendritic processing of AN inputs determines the frequency selectivity and temporal responses characteristic of stellate cells observed in frequency threshold curves PSTHs, and modulation transfer functions.
Author: Quenet B.*, Devaud J. M. Gascuel J., and Masson C.

Laboratoire de Neurobiologie Comparee des Invertebres (L.N.C.I.)
INRA-CNRS (URA 1190)
BP 23 91440 Bures-sur-Yvette, France

Title: IS A CLASSIFICATION OF HONEYBEE ANTENNAL LOBE NEURONES GROWN IN CULTURE POSSIBLE?

Abstract: The study presented here concerns the morphology of honeybee antennal lobe neurones grown in vitro. An important set of such neurons has been described by using morphometric parameters that permit to classified them into three different classes. An attempt to make a correspondence between this classification and the one established in vivo on morphological and functional criteria is proposed.
Author: Subrata Rakshit* and Charles H. Anderson

Dept. of Anatomy and Neurobiology
Washington University School of Medicine
St. Louis, MO, 63110

Title: MAXIMALLY INFORMATIVE FILTER BANKS AND NEURAL ENCODING

Abstract: A multitude of neural responses and representations are observed experimentally and utilized in models. The choice of which representation to use for a particular computational task can be explored utilizing information theory. It can be shown that in order to be maximally informative about their inputs, neural response curves must encode the probability density functions of their inputs. Moreover, the information in the neural outputs must be encoded in a way that makes it immune to noise inherent in spike trains. Hence the selection of the information encoding scheme is influenced by the structure of the statistics of the task space, the signal to noise of neural spike train outputs and the amount of resources that can be devoted to the task. Specific neurobiological examples are examined to show how far one may account for various biological design strategies by assuming that the goal is to robustly encode maximum amount of information at a given cost.
Author: Walter Read* 1 and Valeriy I. Nenov 2

1 Department of Computer Science
California State University
Fresno, CA

2 Division of Neurosurgery, CHS 74-140
UCLA School of Medicine, Los Angeles, CA

Title: A COMPUTATIONAL MODEL OF ATTENTIONAL FOCUS SEARCHLIGHT OF ATTENTION HYPOTHESIS REVISITED

Abstract: The "searchlight of attention" can involve (at least) two separate functions: the relative enhancement of one region and a shift of the enhancement. To study these functions we built a series of computational models of increasing complexity. These models were analyzed and implemented on the Connection Machine for a variety of parameter values. We found that under reasonable assumptions, the model showed strong enhancement of the most prominent area and was able to fade out the foreground and to enhance the background. We have not yet been able to show full general switching between visual areas.
Author: Dr. Maureen E. Rush 1 and Dr. John Rinzel 2

1 Mathematics Department
California State University
Bakersfield, CA 93311

2 Mathematical Research Branch
NIDDK/NIH
Bethesda, MD 20892

Title: THE POTASSIUM A-CURRENT, LOW FIRING RATES, AND REBOUND EXCITATION IN HODGKIN-HUXLEY MODELS

Abstract: The Hodgkin-Huxley equations have been used to describe action potential generation in many excitable cells, and they predict repetitive firing over an interval of constant, applied current values. Moreover, the minimum firing frequency is rather high (> 50 Hz) as a consequence of the subcritical Hopf bifurcation structure of periodic solutions to the equations. In the work of Connor and Stevens, lower firing rates are attributed to the presence of a second potassium current, the transient A-current. We show that a necessary condition to obtain arbitrarily low rates is that the membrane's steady-state, current-voltage relation be non-monotonic, and that this leads to a different (non-Hopf-like) bifurcation structure for the emergent periodic behavior. Motivated by the prevalence of the A-current in many excitable cells, we define a generic A-current and study the onset of periodic orbits with zero frequency and the dependence of this non-Hopf-like bifurcation on A-current parameters. For the data set we consider most typical, we find that IA does not lower the firing frequency arbitrarily close to zero; it does shift the stimulus-frequency curve to a more depolarized interval of applied current, and so effectively reduces the frequency of firing. Finally, we use the method of averaging to show how additional bifurcations emerge along the branch of periodic solutions if the inactivation of IA is slow. This leads to periodic bursting of action potentials where the number of sodium spikes in a burst depends on the maximal conductance of IA-

Many cell types that fire repetitively under constant depolarizing stimulus also show the phenomenon of anode break excitation. The A-current can significantly effect such rebound behavior since it recovers from inactivation during hyperpolarization. The additional opposition provided by IA can suppress excitation after deep hyperpolarization, creating an 'inhibition' threshold above which a cell can no longer fire. Since the model neuron remains excitable after modest hyperpolarization, an excitation window is created. IA controls the size of this window, and there is a maximal conductance value above which all excitation is suppressed.
Author: I. A. Rybak* and J. S. Schwaber

Neural Computation Group
Experimental Station
DuPont Co, E-323
Wilmington, DE 19880-0323

Department of Neuroscience
University of Pennsylvania
Philadelphia, PA 19104

Title: COMPARATIVE MODELING OF NEUROGENESIS OF THE THREE-PHASE RESPIRATORY RHYTHM

Abstract: Models capable of autonomous generation of the respiratory rhythm were developed following Richter's theory of the origin of the respiratory rhythm. In this theory the respiratory cycle consists of three phases: inspiration, post-inspiration and expiration; and the oscillations in the respiratory network are provided by both specific interconnections and individual properties of neurons of several respiratory groups. Several versions of the neural architecture capable of generating the three-phase respiratory rhythm and reproducing the specific activity patterns of real neurons of different respiratory groups have been developed. Model comparisons were made of a simplified spiking, dynamic neuron model and the complete Hodgkin-Huxley type model. Results are analyzed and compared with physiological data. Some predictions are considered.
Author: Sabbatini, R.M.E 1 and Cardoso, S.H. 2

1 Center for Biomedical Informatics of the State University of
Campinas
P. O. Box 6005
Campinas, SP 13081-970, Brazil

2 Laboratory of Psychobiology
Dept. of Psychology and Education
University of Sao Paulo
Ribeirao Preto
SP 14040-000, Brazil

Title: CLASSIFICATION AND QUANTIFICATION OF BEHAVIORAL PATTERNS AND SEQUENCES IN NEUROETHOLOGICAL STUDIES, USING ARTIFICIAL NEURAL NETWORKS

Abstract: The identification, isolation and quantification of animal behavioral patterns and sequences represent essential steps for the analysis of observed behavior using the ethological approach. Many of the difficulties and poor performance associated to conventional techniques can be traced to time-varying transition probabilities between elements of a sequence and the essential non-linear separability of multidimensional patterns. In this paper we present a novel approach to the problem of behavioral pattern classification, using artificial neural networks (ANN). Our aim was to test the feasibility of using ANNs in the task of automatic identification and quantification of complex behavioral patterns, using a input a raw sequence of elementary behavioral items recorded systematically by direct observation, consisting of computer-recorded sequences of aggressive, defensive and escape behavior of brain-stimulated cats. All networks converged to criterion. The present work provides a demonstration of the usefulness and viability of the ANN approach in computational analysis of observed behavioral patterns in neuroethology.
Author: Sabbatini, R.M.E.

Center for Biomedical Informatics
State of University of Campinas
P.O. Box 6005
Campinas, SP 13081-970, Brazil

Title: USING ARTIFICIAL NEURAL NETWORKS TO UNDERSTAND BRAIN FUNCTION: THE ANALYSIS OF NEUROELECTRIC INFORMATION

Abstract: One of the main goals of the neurosciences is to measure and to understand the complex flow of information that takes place in the nervous system. With the recent development of techniques for the simultaneous recording of large number of neurons the informational analysis of recorded spike-train data is crucial to the study of the dynamics of biological neural networks. Artificial neural networks (ANNs) are proving to be very useful as computational tools for this purpose, thus closing the circle of mutual influence between the Neurosciences and Connectionist Science. This paper reviews briefly the state-of-the-art in ANN applications in cell-level neuroelectric signal processing, and discusses new ideas for a more intense bond between connectionist science and computational neurosciences.
Author: Emilio Salinas* and L.F. Abbott

Biology Department and Center for Complex Systems
Brandeis University
Waltham, MA 02254

Title: DECODING VECTORIAL INFORMATION FROM FIRING RATES

Abstract: In many systems the firing rates of a population of neurons are functions of a vectorial quantity. We address the problem of reconstructing the coded vector from measured firing rate data. Several methods are analysed and compared. A new linear method, the optimal linear estimator (OLE), is presented. When neurons have tuning curves that approximate cosines the OLE generates estimates of the vector that are as accurate as those found using complex statistical methods. It also produces more accurate reconstructions using many less neurons than the known vector method. The results point out two general information coding strategies that neural networks might implement.
Author: Samir I. Sayegh* 1, Pulin Sampat 2 and Sattar A. Jaboori 3

1 Department of Physics
2 Department of Computer Sciences
3 Department of Biological Sciences

Purdue University at Fort Wayne
Fort Wayne, IN 46805-1499

Title: ANALYZING THE HIPPOCAMPAL PLACE-CELL PHENOMENON BY MODELING THE CENTRAL VISUAL PATHWAY

Abstract: We have constructed a realistic, yet simple model in GENESIS incorporating the place-cell activity in response to visual cues as processed through the central visual pathway. The CA3 region in the hippocampus contains bursting pyramidal cells that respond to spatial location by processing environmental cues. Our pyramidal cells as constructed are based on the model developed by Roger Traub (1991). Our results show the movement of stimuli induced by environmental cues through the central visual pathway and eventually mapped onto the hippocampus. The model learns the position of the environmental cues via a hebbian mechanism.
Author: Harel Shouval* and Yong Liu

Department of Physics and
Institute for Brain and Neural Systems
Box 1843, Brown University
Providence, R. I., 02912

Title: HOW DOES RETINAL PREPROCESSING EFFECT THE RECEPTIVE FIELDS OF STABILYZED HEBBIEN NEURONS

Abstract: The structure of receptive fields in the visual cortex is believed to be shaped by unsupervised learning. It has been shown that several of the forms of stabilyzed Hebbien (Hebb, 1949) rules are governed by the first principal components(Oja, 1982; ?). In this paper we analyze the form of the principal components of natural images, which have been preprocessed by center surround filters, anoulogus to those found in the retina. An assumption is made that only small circular regions of the images are being used as training patterns. We investigate how the ratio between the size of the receptive fields, and the size of the preprocessing filter, effects the receptive field structure. The derivation relies on results about the correlation function of natural images (Field, 1987), and on the assumption that the correlation function is radially symmetric. Finally the biological relevance of our results is discussed.
Author: Karen A. Sigvardt* 1 and Thelma L. Williams 2

1 Department of Neurology
University of California-Davis
Martinez CA 94553

2 Department of Physiology
St George's Hospital Medical School
London SW17 ORE, UK

Title: SIMULATION AND EXPERIMENTAL CONFIRMATION OF A BOUNDARY LAYER IN THE INTERSEGMENTAL PHASE LAGS ALONG THE LENGTH OF THE LAMPREY SPINAL CORD.

Abstract: An important feature of the swimming motor pattern generated by the lamprey spinal cord is an intersegmental phase delay that is constant along the length of the cord. The lamprey CPG for locomotion has been modeled as a chain of coupled oscillators, within a general mathematical framework (Kopell and Ermentrout 1986, 1988). The analysis predicts that for asymmetric coupling of equally-activated oscillators, the intersegmental phase lag will be uniform along the chain except in a boundary layer at one end. In this presentation we will describe simulations showing this boundary layer at the rostral end of a chain of oscillators in which ascending coupling is dominant, and experimental results that confirm that a boundary layer does exist at the rostral end of an isolated preparation of lamprey spinal cord.
Author: Joseph Sirosh* and Risto Miikkulainen

Department of Computer Sciences
The University of Texas at Austin
Austin, TX 78712

Title: MODELING CORTICAL PLASTICITY BASED ON ADAPTING LATERAL INTERACTION

Abstract: A neural network model called LISSOM for the cooperative self-organization of afferent and lateral connections in cortical maps is applied to modeling cortical plasticity. After self-organization, the LISSOM maps are in dynamic equilibrium with the input, and reorganize like the cortex in response to simulated cortical lesions and intracortical microstimulation. Adapting lateral interactions are shown to form the basis for such plasticity. The model replicates several experimental results computationally and can also account for the dynamics of cortical reorganization.
Author: William Lincoln and Josef Skrzypek*

Machine Perception Laboratory
University of California
Los Angeles, CA 90024

Title: DEPTH FROM TRANSPARENCY

Abstract: When transparency is perceived, psychophysical evidence demonstrates effects on depth relationships between two or more apparently superimposed physical surfaces suggestive of early interaction. Why should there exist low-level visual mechanisms for a seemingly peripheral visual phenomenon such as transparency? Shadows as well as other physical phenomenon cannot create the luminance and depth conditions consistent with perceptual experience, and thus mechanisms for their analysis are not solely responsible for perceptual transparency. We present here a computational model for perceptual transparency supposing no specialized mechanisms. In our model depth from transparency results from local interactions of monocular occlusion cues and stereoscopic depth. End-stopped cells might provide the physiological mechanism sensitive to opaque occlusion. Simulation results suggest the pattern of activity of hypothetical end-stopped cells determines the depth interactions due to transparency. End-stop cells signaling occlusion interact with a population encoding of disparity consistent with neurophysiology and psychophysics in a computational model that explains many experimental results.
Author: D. K. Smetters* and S. B. Nelson

Department of Brain and Cognitive Science
43 Carleton St., E25-618
MIT
Cambridge, MA 02139

Title: ELECTROTONIC STRUCTURE AND SYNAPTIC VARIABILITY IN CORTICAL NEURONS

Abstract: Compartmental models of reconstructed cortical neurons were used to assess the relative contributions of electrotonic filtering, synaptic parameters and recording characteristics on the distribution of synaptic responses measured at the soma from synapses located throughout the dendritic tree. The distribution of peak amplitudes in simulated current clamp was very narrow, except for extremely distal synapses. In voltage clamp, space clamp errors and dendritic filtering increase variability, but this is greatly reduced by series resistance. This suggests that while some of the synaptic variability measured experimentally is due to cable filtering, much of it results from intrinsic variability between synapses.
Author: D. C. Somers*, S. B. Nelson, and M. Sur

Department of Brain and Cognitive Science
43 Carleton St., E25-618
MIT
Cambridge, MA 02139

Title: AN EMERGENT MODEL OF VISUAL CORTICAL ORIENTATION SELECTIVITY

Abstract: We demonstrate a cortical circuit based on known anatomical details and cellular properties which can achieve sharp orientation tuning despite poorly tuned thalamocortical input. Sharp tuning arises emergently from excitatory interactions between similarly tuned neurons, which also receive broadly tuned inhibition. This model accounts for intracellular recordings and pharmacological blockade studies, the results of which had appeared to conflict over the role of inhibition. We suggest that inhibition acts non-specifically to maintain the selectivity of individual neurons, but that it is critical at the columnar level for balancing the strong intracortical excitation.
Author: Martin Stemmler*, Marius Usher, and Christof Koch

Computation and Neural Systems, 139-74
California Institute of Technology
Pasadena, CA 91125

Title: OSCILLATORY FIELD POTENTIALS WITH IRREGULAR SINGLE CELL DISCHARGE

Abstract: While cortical oscillations are robust and commonly found in local
field potential measurements (Gray et al., 1990; Eckhorn et al., 1993), they are much less evident in single spike trains recorded from behaving monkeys (Bair et al.,l994). We show that a simple computational neural model with biologically plausible lateral connectivity can explain such a discrepancy. Analysis of the model's power spectra reveals a prominent peak around 30-50 Hz in the local field potential, defined as the summed spiking activity of a local population. At the same time, oscillatory peaks are completely absent in the spectra of single cell spike trains. Instead, the discharge pattern of single cells is highly irregular, as it is in vivo (Softky and Koch, 1993); as a consequence, single-cell power spectra replicate the standard spectral shape seen in cortical cells (Bair et. al, 1994 ).
Author: V. Kowtha 1, P. Satyanarayana 2 , R. Granger 3 and D. Stenger* 4

1 Center for Bio/Molecular Science and Engineering, Code 6900
Naval Research Laboratory
Washington, DC 20375

2 Applied Physics Division
Science Applicaitons International Corporation
1710 Goodrich Drive
Mclean, VA 22102

3 Center for Neurobiology of Learning and Memory
University of California, Irvine, CA 92717

4 Center for Bio/Molecular Science and Engineering, Code 6900
Naval Research Laboratory
Washington, DC 20375

Title: EFFECTS OF RANDOM NOISE ON CLASSIFICATION BY A PIRIFORM HIERARCHICAL NEURAL NETWORK

Abstract: Fundamental questions exist about the ability of biological and artificial neural networks to perform classification when presented with increasingly noisy input patterns. We compared a piriform hierarchical clustering (PHC) algorithm, modeled after the rat olfactory cortex, with conventional back propagation algorithms. PHC can be trained on patterns with combined levels (0, 1, 5, and 10%) of Gaussain noise to allow roughly equivalent performance at arbitrary noise levels between 0 and 10%. We examine the mechanisms underlying this property, and suggest possible implications for modeling of biological neural networks that must operate on input patterns with varying signal-to-noise ratio.
Author: Edward Stern*, Charles Wilson, and Anthony Kincaid

Dept. of Anatomy and Neurobiology
University of Tennessee Medical School
855 Monroe Ave.
Memphis, TN 38163

Title: INFORMATION PROCESSING PROPERTIES OF CORTICOSTRIATAL AND NEOSTRIATAL NEURONS

Abstract: We analyzed the membrane bistability of neostriatal and corticostriatal neurons using dwell-time histograms. The variance of the depolarized state was greater than that of the hyperpolarized state, reflecting synaptic noise. The bistability of the corticostriatal cells is less clearly defined, and has a smaller dynamic range than that of the neostriatal neurons. The analysis showed a high likelihood that the converging synaptic input to the cortical and corticostriatal neurons has a large stochastic component. The onset of the input needed to generate the depolarized state occurs with similar distributions in both neuronal classes. The differences can be explained by the different distributions of channels in the respective neuronal cell membranes.
Author: Edward Stern* 1, Ad Aertsen 2, Eilon Vaadia 3, Shaul Hochstein 4

Dept. of Anatomy and Neurobioloy
University of Tennessee Medical School,
855 Monroe Ave.
Memphis, TN 38163 USA

2 Dept. of Neurobiology, Weizmann Inst. of Science, Rehovot, Israel,
and Inst. fuer Neuroinformatik, Ruhr-University, Bochum, Germany.

3 Physiology Dept and Center for Neural Computation
Hebrew University, Jerusalem, Israel

4 Neurobiology Dept and Center for Neural Computation
Hebrew University, Jerusalem, Israel

Title: REDUCTION OF STIMULUS AMBIGUITY: FUNCTIONAL CONNECTIVITY AND CORTICAL OPERATIONS

Abstract: Multidimensional receptive fields were used for the description and comparison of the information-processing characteristics of single neurons and neuronal assemblies. The tuning of the response of an assembly of simultaneously recorded neurons was not predicted by the individual responses of the component neurons. The bandwidth of the multineuron response was significantly narrower than that predicted from the tuning of the individual neurons' receptive fields. The coincident firing of two neurons therefore carries more information than the two neurons measured individually. Neuronal coincident activity is therefore a dynamic. stimulus- and time-dependent phenomenon. We propose that this may be a major computational mechanism of primary sensory cortex.
Author: Michael Stiber* 1 , Li Yan 1 and Jose P. Segundo 2

1 Department of Computer Science
The Hong Kong University of Science and Technology
Clear Water Bay, Kowloon, Hong Kong

Jose P. Segundo
Department of Anatomy and Cell Biology
and Brain Research Institute
University of California
Los Angeles, California 90024

Title: SYNAPTIC CODING OF NONSTATIONARY SPIKE TRAINS

Abstract: Spike-producing neurons produce complex responses to stationary input trains. These responses have been described using techniques from the field of nonlineary dynamics, and are typical of those from periodically perturbed nonlinear oscillators.

Here we are concerned with the effects of nonstationary input trains. We present recent simulation results, largely in agreement with experimental results on a living preparation, emphasizing the relationships between stationary and nonstationary behaviors. The implications for synaptic coding are considered. We suggest that the viewpoint of a neuron as a nonlineary dynamical system has important contributions to make to our understanding of neural computation.
Author: A. Surkis 1, B. Taylor 1, C. S. Peskin 2, and C. S. Leonard 1

Center for Neural Science, New York University,
6 Washington Place, New York, NY 10003

Courant Institute of Mathematical Sciences, New York University,
251 Mercer Street, New York, NY 10003

Title: CALCULATION OF PASSIVE MEMBRANE PROPERTIES FOR ARBITRARY DENDRITIC GEOMETRY OF LATERODORSAL TEGMENTAL (LDT) NEURONS IN VITRO.

Abstract: A numerical solution of the cable equation was used to extract the membrane parameters from guinea pig cells that were studied with current injection in a brain slice preparation. Reconstructions of cells provided the geometries used in the numerical solution. Cable parameters were chosen which provided the best fit to experimental data, where the error was calculated directly between the recorded voltage traces and the cable equation solution. In searching the parameter space, intracellular resistivity was held fixed, while membrane conductance and capacitance varied. Fits were also done for the case in which dendritic and somatic membrane conductance varied independently.
Author: Jeffrey P. Sutton* 1, and James A. Anderson 2

1 Department of Psychiatry
Harvard Medical School
MGH Bldg 149, 13th Street
Charlestown, MA 02129
and
Cognitive Sciences
E25-201, Massachusetts Institute of Technology
Cambridge, MA 02139
sutton@ai.mit.edu

2 Department of Cognitive and Linguistic Sciences
Brown University, Providence, RI 02912

Title: COMPUTATIONAL AND NEUROBIOLOGICAL FEATURES OF A NETWORK OF NETWORKS

Abstract: A recurrent theme in neurobiology is that modules of neurons process local information. When modules are modeled, the approximation is almost always made that the overall average activity level is the critical measure. We propose a model wherein the communication between the local networks is vector valued rather than a scalar quantity representing average activity. We term our model system a "network of networks." In this contribution, we review and integrate the computational and neurobiological aspects of the model and suggest ways for implementing the approach to problems in computer science, neuroscience and neurology.
Author: David C. Tam* 1 and Michelle A. Fitzurka 2

1 Center for Network Neuroscience
Dept. of Biological Sciences
University of North Texas
Denton, TX 76203

2 Dept. of Physics
Catholic University
Washington, DC 20064

Title: A NEW SPIKE TRAIN ANALYSIS FOR DETECTING TRENDS IN FIRING PATTERNS IN NEURONS

Abstract: We have developed a new spike train analysis method for detecting trends in firing patterns in neurons based on a statistical measure of serially dependent firing probabilities. We introduce the serial interspike interval difference (SISID) scatter plot to display these changes in firing probability. This technique reveals a higher order serial dependency by showing how such "trends" evolve in time within a spike train. SISID analysis establishes the statistics based on the difference between two adjacent interspike intervals (ISIs). Typically, joint interspike interval (JISI) analysis estimates the conditional firing probabilities at consecutive ISIs by plotting (tx,ty). Similarly, SISID plots the difference between two adjacent ISI tuples (Ætx, Æty). The distribution of a tuple in the SISID scatter plot indicates the firing trend. For example, points within the first quadrant show that the neuron fires with increasing ISIs, while those in the third quadrant denote the opposite. Furthermore, by connecting adjacent points in the scatter plot, the evolution of the firing trend can be revealed. This allows us to determine how firing patterns change, thus enabling the detection of characteristic firing patterns or burst behaviors for that neuron.
Author: Robert Thau

MIT Dept. of Brain and Cognitive Sciences
MIT, room NE43-711
Cambridge, MA, 02139

Title: VISUAL SEGMENTATION AND FEATURE BINDING WITHOUT SYNCHRONIZATION

Abstract: There has recently been a sizable amount of work on visual segmentation in neural networks. Typically, these have functioned by "tagging" the outputs of units, by oscillatory phase or some similar mechanism.

The network described here exploits constraints on the segmentation task to solve it with units whose outputs have no special features for binding or tagging at all, using a process analogous to constraint propagation among Waltz labels for the scene. Possible physiological correlates of the network's mechanism in primate visual cortex are discussed.
Author: Frederic Theunissen* and Gwen Jacobs

Dept. of Molecular and Cell Biology Neurobiology
Division. LSA 195. Berkeley, CA 94720

Title: THE ROLE OF FUNCTIONAL MAPS: FACILITATING CONNECTIVITY AND OPTIMIZING INFORMATION TRANSMISSION

Abstract: We postulate that functional sensory maps are generated to facilitate the connectivity between the afferent neurons and the output neurons in a manner that optimizes the transfer of information across the synaptic interface. In our theory, the connections between the output neurons and afferents are determined by simple rules which only invoke the spatial location of the input dendrites of the output neurons. Within our theory, we derived a mathematical formalism to 1) calculate the efficiency of a functional map in transmitting the maximum amount of information and 2) obtain a measure how simple the connectivity rules between the map and the output neurons need to be to achieve such an efficiency. High map efficiencies and simple connectivity rules would give good supporting evidence for our theory. The formalism is applied to the wind direction map of the cercal system of the cricket with astonishingly good results.
Author: Philip S. Ulinski* and Mathew T. Calef

Department of Organismal Biology and Anatomy
University of Chicago
1025 E. 57th Street, Chicago, IL 60637

Title: SYNAPTIC MECHANISMS UNDERLYING INTENSITY-RESPONSE PROFILES IN CORTICAL NEURONS.

Abstract: Responses of neurons in visual cortex to lights of different intensities were studied using an in vitro preparation of turtle whole-brain and eyes and a compartmental model of cortical microcircuitry. Pyramidal neurons show GABAA-mediated IPSPs mediated via feedforward inhibition from layer 1 stellate cells in the dark. The frequency of GABAA-mediated IPSPs increases and EPSPs mediated by geniculate afferents appear at moderate levels of light intensity. G A B A A- and GABAB-mediated inhibition mediated by feedback from stellate and horizontal cells limits the number of action potentials produced by the pyramidal cell at higher levels of light intensity.
Author: Jaap van Pelt 1 and Andreas Schierwagen 2

1 Netherlands Institute for Brain Research
Meibergdreef 33
1105 AZ Amsterdam, The Netherlands

2 University of Leipzig
Dept. of Informatics
Augustusplatz 10/11
O-7010 Leipzig, Germany

Title: DEPENDENCE OF DENDRITIC ELECTROTONIC EXTENT UPON BRANCHING PATTERN TOPOLOGY

Abstract: The effect of branching pattern topology on the dendritic electrotonic extent is investigated in a systematic statistical way. Sets of dendritic branching patterns with a realistic topological variability are produced using a stochastic model for dendritic outgrowth.

The present study demonstrates not only that the electrotonic extent of a dendrite depends strongly on its topological structure, but also that the branch-power, used to calculate the segment diameters, plays a critical role herein. Additionally, the three different measures, used to express dendritic electrotonic extent, give inconsistent outcomes when the branch power differs from the value of 1.5 or when the trees are very asymmetric.
Author: C.A. van Vreeswijk* 1, L.F. Abbott 1, and G.B. Ermentrout 2

1 Department of Physics and Center for Complex Systems
Brandeis University
Waltham, MA 02254

2 Department of Mathematics
University of Pittsburg
Pittsburg, PA 15260

Title: WHEN INHIBITION NOT EXCITATION SYNCHRONIZES NEURAL FIRING

Abstract: It is commonly assumed that excitatory synapses tend to synchronize neurons while inhibition pushes neurons toward anti-synchrony. However the opposite has been observed in some models. In this study we show that such 'reversed' behaviour is the rule rather than the exception. We consider circuits of two identical neurons with fast, well-separated action potentials and with identical synapses having time-constants exceeding the spike width. We consider integrate-and-fire models, as well as models using phase-coupling. We find that only inhibition completely synchronizes the cells. Excitation often produces anti-synchronous firing. These results were confirmed by computer simulations using Hodgkin-Huxley model neurons.
Author: J. - F. Vibert, D. Lambolez, K. Pakdaman, and N. Azmy

B3E, URBB INSERM
U263 Faculte de Medecine Saint-Antoine.
27 rue Chaligny 75571 Paris Cedex 12, France

Title: XNBC: A SIMULATION TOOL FOR NEUROBIOLOGISTS.

Abstract: XNBC allows the simulation of interacting biological neural networks. It is designed for computer naive neuroscientists. XNBC provides two simulation levels: one based on a leaky integrator model of neuron dynamics and another based on the Hodgkin-Huxley formalism describing ionic channel conductance dynamics. Simulated biological networks can be made of several interconnected neural clusters, which may receive external inputs even from experimental spike train recordings. Network architecture and neuron parameters are specified and fine tuned using graphic editors under visual control. Parameters can be modified on the fly during simulations in order to reproduce experimental or pathological conditions. Animation of the neurons' membrane potential spatio-temporal evolution, of spikes travelling along axons and the networks' global activity can be visualized on moving displays. XNBC provides the complete collection of analysis tools used by neuroscientists, to study the network behavior, at both global and unitary levels, in both frequency and temporal domains. XNBC runs on Unix, is menu driven with a user-friendly XWindow/Motif interface and produces high quality color PostScript graphic outputs. abstract document.
Author: Tom Wadden*1, Jeanette Hellgren-Kotaleski 2, Anders Lansner 1, and Sten Grillner 2

1 Department of Numerical Analysis and Computing Science
Royal Institute of Technology
Stockholm, Sweden

2 Nobel Institute for Neurophysiology
Karolinska Institutet
Stockholm, Sweden

Title: SIMULATIONS OF THE INTERSEGMENTAL COORDINATION USING A CONTINUOUS NETWORK MODEL

Abstract: Swimming in the lamprey involves the coordination of alternating burst activity between the left and right sides of each spinal segment, with a frequency from .25 to 10 Hz. Rostrocaudal time delays between burst onset in each segment produce a laterally directed traveling wave which pushes the animal forward through the water. A reversed direction of the wave results in backward swimming. In our model we use a compartmental based neuron employing a Hodgkin-Huxley type formalism, with the synaptic connectivity compatible with experimental data. The result is a continuous network model for intersegmental coordination that is capable of producing stable forward, backward and narrow swimming.
Author: J. Thomas Westbrook* 1 and Daryl R. Kipke 2
Arizona State University
Bioengineering Program
Tempe, AZ 85287-6006

Title: INCREASING DYNAMIC RANGE IN A COCHLEA MODEL USING HIGH, MEDIUM, AND LOW SPONTANEOUS RATE FIBERS

Abstract: We present a composite auditory periphery model that simulates cochlear mechanics, hair cell physiology, and auditory-nerve fiber physiology. It includes three separate populations of auditory nerve fibers having low, medium, and high spontaneous firing rates. We have implemented it on a Connection Machine and a UNIX workstation and have varified it using tones, amplitude-modulated tones, noise, and speech. In each case the model output compares well with published physiological data. The model operates over the entire dynamic range and frequency range of human hearing.
Author: Charles Wilson

Department of Anatomy and Neurobiology
University of Tennessee
Memphis 875 Monroe Ave.
Memphis, TN 38l63

Title: OUTWARD RECTIFICATION IN DENDRITES: COMPUTER SIMULATIONS OF ITS EFFECTS ON SYNAPTIC TRANSMISSION

Abstract: Computer simulations were used to study the effect of voltage-gated potassium conductances on summation of excitatory synaptic potentials in a dendritic neuron. They acted to counteract spatial gradients of potential over the cell membrane. They isolated and suppressed spatially clustered, temporally asychronous inputs. Spatially distributed input saturated at membrane potentials well below the synaptic reversal potential.
Author: Winslow

Biomedical Engineering
University of Toronto
Toronto, Ontario M5S-3B9

Title: EFFECTS OFENDOGENOUS CALCIUM BUFFERS AND VESICLE LOCATION ON VESICLE RELEASE DYNAMICS AND NEUROTRANMI l l~R RELEASE

Abstract: In presynpatic axoplasmic regions, calcium enters via voltage gated channels in active zones. This rapid influx of calcium interacts with trigger or release (T/R) molecules to cause transmitter vesicle fusion with membrane. Using reaction-diffusion PDEs for concentrations of calcium, nondiffusable buffer and product, diffusable buffer and product, calcium concentrations were calculated. Differences in concentration of nondiffusable buffer has considerable effects on fast dynamics of calcium concentration. These calcium dynamics have profound influence on the T/R mechanisms for vesicle fusion with membrane. Location of vesicles with respect to channels and closeness of active zones have a large influence on release dynamics.
Author: F. Wolf*, K. Pawelzik, T. Geisel

Institut fur Theoretische Physik
Universitat Frankfurt
60054 Frankfurt/M., Germany

Title: EMERGENCE OF LONG RANGE ORDER IN MAPS OF ORIENTATION PREFERENCE

Abstract: The formation of feature-maps in the developing visual cortex has become a central system for the study of cooperative phenomena in large neural networks, both theoretically and experimentally. With advanced optical imaging techniques it has now become possible to monitor the activity of neural populations synchronously with a high spatial resolution. This kind of measurement should finally enable a comparison of experiment and mathematical theory in quantitative detail. In the adult visual cortex the pattern of preferred orientations exhibits a particular complex spatial organization, characterized by a large number of point defects (pinwheels). Because the orientation preference map in adult cats is close to optimally smooth, the structure of the map can be predicted from the knowledge of position and chirality of its defects. Moreover in this system optimal smoothness results in a particular kind of long range order, that allows to predict preferred orientation across cortical distance. In this contribution we show that this global spatial coherence cannot be achieved easily during the initial phase of development, in which the pattern of preferred orientations arises via an instability mechanism. Therefore we propose a two stage model for the process of map formation which explains the establishment of long range order in the visual cortex. We predict the occurrence of a second phase in the process of map formation, during which cells cooperate effectively over large cortical distances. Furthermore we show how this process could be observed directly in optical imaging experiments 1.
Author: Shih-Cheng Yen*, Paul Sajda, and Leif Finkel 3

Department of Bioengineering and Institute for Neurological Sciences
220 S. 33rd St.
University of Pennsylvania
Philadelphia, PA 19104-6392

Title: FACE RECOGNITION BY PDP AND RADIAL BASIS FUNCTION NETWORKS: COMPARISONS AND INSIGHTS

Abstract: Despite a number of proposed neuropsychological and computational models, there is no accepted explanation for human face recognition abilities. We investigated the representations developed in both PDP and Radial Basis function networks presented with a large database of faces. Both networks achieved performances above 90% in gender classification tasks. Network representations were analyzed using a number of techniques including examination of connection weights, network inversion, ablation and modification of the image, and Wiener kernal - reverse correlation techniques. Comparison of the networks reveals a template-based strategy that combines statistical decision making with proto-type exemplars.
Author: Alan L. Yuille* 1 and Stelios M. Smirnakis 2

1 Harvard University
G12e Pierce Hall
Division of Applied Sciences
29 Oxford Street
Cambridge, MA 02138

2 Division of Applied Sciences
Harvard University
Cambridge, MA 02138

Title: NEURAL IMPLEMENTATION OF BAYESIAN VISION THEORIES BY UNSUPERVISED LEARNING

Abstract: It is often claimed that Bayesian theories of vision require time consuming, and biologically implausible, relaxation algorithms. We show that, on the contrary, Bayesian theories can be implemented by feedforward networks, multilayer perceptrons, where the weights of the network are trained by unsupervised learning using a novel variant of backpropagation. Both multilayer perceptrons and backpropagation are of questionable realism but our approach can be generalized to more biologically plausible models. We illustrate our theory on an example of image segmentation. Such unsupervised learning might have a role in the development of the visual system.