Neuroscience: From Neural Networks to Artificial Intelligence: Proceedings of a U.S.-Mexico Seminar held in the city of Xalapa in the state of Veracruz on December 9–11, 1991

587 Pages·1993·14.334 MB·English

by  José Bargas, Elvira Galarraga, D. James Surmeier (auth.), Pablo Rudomin, Michael A. Arbib, Francisco Cervantes-Pérez, Ranulfo Romo (eds.)

Research Notes in Neural Computing
Managing Editor
Bart KosKo
Editorial Board
S. Amari M. A. Arbib C. von der Malsburg
Advisory Board
Y. Abu-Mostafa A. G. Barto E. Bienenstock J. D. Cowan
M. Cynader W. Freeman G. Gross
U. an der Heiden M. Hirsch T. Kohonen J. W. Moore
L. Optic an A. I. Selverston R. Shapley B. Soffer
P. Treleaven W. von Seelen B. Widrow S. Zucker
Pablo Rudomin Michael A. Arbib
Francisco Cervantes-Perez Ranulfo Romo
Editors
Neuroscience:
From Neural Networks
to Artificial Intelligence
Proceedings of a U.S.-Mexico Seminar
held in the city of Xalapa in the state of Veracruz
on December 9-11, 1991
Springer-Verlag
Berlin Heidelberg New York
London Paris Tokyo
Hong Kong Barcelona
Budapest
Michael A. Arbib Francisco Cervantes-Perez
Center for Neural Engineering Centro de Instrumentos
University of Southern California Universidad Nacional Aut6noma de Mexico
Los Angeles, CA 90089-2520 Apartado Postal 70-183
USA Mexico D.F., CP 04510
Mexico
Ranulfo Romo
Instituto de Fisiologfa Celular Pablo Rudomin
UNAM Centro de Investigaci6n
Mexico D.F. y de Estudios Avanzados del I.P.N.
Mexico Apartado Postal 14-740
Mexico 07000 D.F.
Mexico
Managing Editor
Bart Kosko
Engineering Image Processing Institute
University of Southern California
University Park
Los Angeles, CA 90089-0782
USA
ISBN-13:978-3-540-56501-7 e-ISBN-13:978-3-642-78102-5
DOl: 10.1007/978-3-642-78102-5
Library of Congress Cataloging-in-Publication Data
Neuroscience: from neural networks to artificial intelligence: proceedings of a U.S.
Mexico seminar held in the city of Xalapa in the state of Veracruz on December 9-11,
1991/ edited by Pablo Rudomin ... ret al.l.
p. cm. - (Research notes in neural computing; v. 4)
Includes bibliographical references and index.
ISBN-13:978-3-540-56501-7
I. Neurosciences-Congresses. 2. Neural networks (Computer science )-Congresses.
3. Artificial intelligence-Congresses.
I. Rudomin, Pablo, II. Series. QP35l.N4317 1993 006.3-dc20 93-18575 CIP
This work is subject to copyright. All rights are reserved, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustra-
tions, recitation, broadcasting, reproduction on microfilms or in any other way, and storage
in data banks. Duplication of this publication or parts thereof is permitted only under the
provisions of the German Copyright Law of September 9, 1965, in its current version, and
permission for use must always be obtained from Springer-Verlag. Violations are liable for
prosecution under the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1993
The use of registered names, trademarks, etc. in this publication does not imply, even in the
absence of a specific statement, that such names are exempt from the relevant protective
laws and regulations and therefore free for general use.
Typesetting: camera-ready by authors
33/3140 -543 2 1 0 -Printed on acid-free paper
Introduction
The Central Nervous System can be considered as an aggregate of neurons
specialized in both the transmission and transformation of information.
Information can be used for many purposes, but probably the most important one is
to generate a representation of the "external" world that allows the organism to
react properly to changes in its external environment. These functions range from
such basic ones as detection of changes that may lead to tissue damage and eventual
destruction of the organism and the implementation of avoidance reactions, to
more elaborate representations of the external world implying recognition of
shapes, sounds and textures as the basis of planned action or even reflection. Some
of these functions confer a clear survival advantage to the organism (prey or mate
recognition, escape reactions, etc.). Others can be considered as an essential part of
cognitive processes that contribute, to varying degrees, to the development of
individuality and self-consciousness.
How can we hope to understand the complexity inherent in this range of
functionalities? One of the distinguishing features of the last two decades has been
the availability of computational power that has impacted many areas of science. In
neurophysiology, computation is used for experiment control, data analysis and for
the construction of models that simulate particular systems. Analysis of the
behavior of neuronal networks has transcended the limits of neuroscience and is
now a discipline in itself, with potential applications both in the neural sciences and
in computing sciences. Pattern recognition, robotics and computer vision constitute
interesting lines of research which often utilize concepts arising from the
neurosciences.
However, a limitation of the explosive growth of these closely related disciplines
is the decreased communication between the different groups, a decrease potentiated
by increasing specialization both in language and methods. Within this general
framework it seems necessary to make some effort to cross boundaries and to have
active scientists from different backgrounds and disciplines explain the basic
principles which guide their investigations, with the hope that this will lay the basis
for a future collaborative effort.
Reflecting on this, a group of Mexican scientists determined that the Mexican
scientific community was ripe for an integrative activity of this kind. It was further
VI
agreed that this meeting would have the greatest impact if it were organized in
the form of a U.S.-Mexico Seminar since this would provide a broader range of
disciplinary expertise and ensure the international impact of the proceedings. It
should be noted that current work in artificial intelligence has a wide span, from
computer vision and robotics to natural language processing and symbolic problem
solving. Similarly, neuroscience spans from the finest details of neurochemistry to
the findings of the neurological clinic. It was felt that, in each case, there were areas
that were not yet ripe for dialog. In the end, we picked sensory perception, motor
control and learning in the field of artificial intelligence, and the study of synaptic
interactions and the neurophysiology of defined neural circuitry from neuroscience,
as the focus for the meeting, and designed the program to ensure a rich dialog
between experts in these two areas. Our planning committee discussed the one
other area of possible interaction between artiticial intelligence and neuroscience,
namely that between modeling of cognitive processes and cognitive neuroscience,
but we concluded that the Mexican scientific community did not yet have enough
experts in this area to justify its inclusion in a U.S.-Mexico Seminar. On the
contrary, there was no problem in identifying a critical mass of distinguished
researchers from both the U.S. and Mexico for all the topics we have chosen for the
proposed program.
The meeting took place in the city of Xalapa in the state of Veracruz on December
9 - 11, 1991, and brought together active neurophysiologists with experts in neuronal
network analysis and artificial intelligence to further the integration of their
disciplines by reviewing briefly the state of the art in their fields, defining which are
the basic principles and concepts in which their disciplines are founded, and
providing some perspective on the main problems whose solution will signify an
important advance in knowledge. The meeting was held in the archaeological
museum of Xalapa, and the record of 5000 years of Mexican cultures provided a
stimulating backdrop to a meeting which saw an unusually successful level of
interdisciplinary discussion. We have selected about a quarter of that discussion to
close the volume, to share with our readers some sense of the lively intellectual
exchange which united scientists from very different disciplines in their quest for
mutual understanding. It is our hope that the present volume will further this
exchange. Based on our experience at the meeting, we have rearranged the order of
the papers to provide a more coherent flow of ideas, forming six groups: I. Scales of
m.
Analysis; II. Processing of sensory information; Visual Processing; IV. Learning
and Knowledge Representation; V. Neuronal systems for motor integration; and VI.
VII
Robotics and Control. To this we have added a new Section VII, a Concluding
Perspective which combines an essay on "Methodological Considerations in
Cognitive Science" with a summary, "Viewpoints and Controversies", which
preserves something of the liveliness (but not the length!) of the discussions at the
Xalapa meeting. Each section is preceded by a short introduction, and the reader
may wish to turn to these for a more detailed overview of what the book has to offer
as it explores, at different scales of analysis, the transition from sensory to motor
systems, via the central processes which are continually updated by the processes of
learning.
We express our thanks to the other members of the Program Committee - Pablo
Noriega, Ofelia Cervantes, and Hugo Arechiga - for their help in the planning and
organization of the meeting. We would also like to thank the institutions who
sponsored the Xalapa meeting: Academia de la Investigacion Cientifica (AIC),
Consejo Nacional de Ciencia y Tecnologia (CONACYT), and Instituto Nacional de
Estadistica, Geografia e Informatica (INEGI) in Mexico; the National Science
Foundation (NSF) in the United States; and UNESCO and Red Latinoamericana de
Biologia (RLB) internationally. The Laboratorio Nacional de Informatica Avanzada
(LANIA) and Gobierno del Estado de Veracruz provided the local support for the
meeting in Xalapa and did much to create a most productive and convivial
atmosphere for scientific exchange.
July, 1992 Pablo Rudomin
Mexico, D.F., and Los Angeles Michael A. Arbib
Francisco Cervantes-Perez
Ranulfo Romo
Table of Contents
I. Scales of Analysis......................................................................................... 1
Neuronal networks of the mammalian brain have functionally different classes
of neurons: Suggestions for a taxonomy of membrane ionic conductances
J. Bargas; E. Galarraga; D.J. Surmeier ............................................................. 3
Electrical coupling in networks containing oscillators
E. Marder; L.F. Abbott; A.A. Sharp; N. Kopell ............................................. 33
Dynamical approach to collective brain
M. Zak ........................................................................................................... 43
Schema-theoretic models of arm, hand, and eye movements
M.A. Arbib ...................................... ;. ............................................................ 61
Cooperative distributed problem solving between (and within)
intelligent agents
E.H. Durfee .................................................................................................... 84
II. Processing of Sensory Information ............................................................ 99
Spinal processing of impulse trains from sensory receptors
L.M. Mendell; H.R. Koerber ...................................................................... 103
Central control of sensory information
P. Rudomin ................................................................................................. 116
Parallel and serial processing in the somatosensory system
J.H. Kaas .................................................................................................... 136
Cortical representation of touch
R. Romo; S. Rufz; P. Crespo ...................................................................... 154
An introduction to human haptic exploration and recognition of
objects for neuroscience and AI
S. J. Lederman; R. L. Klatzky .................................................................... 171
Common principles in auditory and visual processing
S.A. Shamma.............................................................................................. 189
III. Visual Processing..................................................................................... 206
Neuronal substrate of ligth-induced attraction and withdrawal in crayfish:
A case of behavioral selection
F. Fernandez de Miguel; H. Arechiga ......................................................... 209
Neural and psychophysical models of chromatic and achromatic
visual processes
E. Martfnez-Uriegas.................................................................................... 231
Computational vision: A probabilistic view of the multi-module paradigm
J.L. Marroquin ............................................................................................ 252
State of the art in image processing
M. Lee; e.H. Anderson; R. J. Weidner.. ..................................................... 267
Shape recognition in mind, brain, and machine
I. Biederman; J.E. Hummel; E.E. Cooper; P.e. Gerhardstein ...................... 282
x
IV. Learning And Knowledge Representation ............................................. 294
Contrasting properties of NMDA-dependent and NMDA-independent
forms of LTP in hippocampal pyramidal cells
R.A. Nicoll; R.A. Zalutsky ........... .... ................................................ .......... 298
Kindling
A. Fernandez-Guardiola; R. Gutierrez; A. Martinez; R. Fernandez-Mas ..... 312
Learning automata: An alternative to artificial neural networks
A. Sanchez Aguilar ..................................................................................... 326
Learning, from a logical point of view
P. Noriega. .................................................................................................. 340
Knowledge representation for speech processing
O. Cervantes ............................................................................................... 359
Data management and inference strategies in a human gait pathology
expert system
G.A. Bekey ................................................................................................. 371
V. Neuronal Systems For Motor Integration ................................................ 385
Entrainment of the spinal neuronal network generating locomotion
G. Viana di Prisco; P. Wallen; S. Grillner.. ................................................. 388
Cortical representation of intended movements
A.P. Georgopoulos ..................................................................................... 398
Saccadic and fixation sytems of oculomotor control in monkey superior
colliculus
R. H. Wurtz; D.P. Munoz ........................................................................... 413
Modulatory effects on prey-recognition in amphibia: A theoretical
experimental study
F. Cervantes-Perez; A. Herrera-Becerra; M. Garcia-Ruiz ........................... 426
VI. Robotics And ControL ............................................................................ 450
Outline for a theory of motor behavior: Involving cooperative actions
of the cerebellum, basal ganglia, and cerebral cortex
J.C. Houk; S.P. Wise .................................................................................. 452
Neural networks and adaptive control
A. G. Barto; V. Gullapalli... ........................................................................ 471
Robustness issues in robot manipulators
C. Verde ...................................................................................................... 494
Symbolic planning versus neural control in robots
C. Torras Genis .......................................................................................... 509
Divine inheritance vs. experience in the world:
Where does the knowledge base come from?
E.M. Riseman; A. R. Hanson ...................................................................... 524
VII. A Concluding Perspective .... .... .... .......................................................... 532
Methodological considerations in Cognitive Science
N. Lara; F. Cervantes-Perez ........................................................................ 533
Viewpoints and controversies ..................................................................... 546
I. Scales of Analysis
The study of how the architectonic characteristics of Central Nervous Systems of
different animals might underlie their different physiological properties, which in
turn may be postulated to subserve behavioral responses, spans from detailed
studies of the ionic conductances of the membrane of a single neuron to
neuroethological experiments in freely moving animals, passing through the
analysis of the performance shown by small and large neural networks. In the first
part of this section we have three papers in this stream.
Jose Bargas, Elvira Galarraga, and James Surmeier provide the reader with a feel
for the richness and subtlety of processing that may occur within a single neuron.
They present a functional taxonomy of membrane ionic conductances and then
explain how these conductances modulate the activation dynamics produced by
specific input signals on a single neuron, as well as how these conductances
underlie various neuronal properties. Persistent or slowly inactivating inward
currents produce sustained steady firing, bistable properties, reverse electrotonic
decay of synaptic potentials, dendritic or high threshold spikes, slow depolarizing
potentials and after-potentials, and neurotransmitter release. This becomes very
important if we take into account that in physiology it is assumed that part of the
information is coded in the neuronal response frequency.
Eve Marder, L.F. Abbott, Andrew Sharp, and Nancy Kopell then show how small
networks of such neurons can exhibit intricate dynamics which are subject to a
range of chemical modulations. Following a hybrid experimental-theoretical
approach, analyze the activation dynamics of a small neural network, and show
how an oscillatory network, considered as part of a bigger network and electrically
coupled to other circuits with different membrane properties, enhances the variety
of dynamic behaviors that may be displayed by the overall network. Through
theoretical analysis, they found that the frequency of a network containing an
oscillator can be either faster or slower than the frequency of the oscillator, the
detailed properties of the oscillator being critically important to control the
frequency of the network. This theoretical result guided an experiment in the
stomatogastric ganglion where the authors modulated the oscillator response
frequency when coupled to a non-oscillatory neuron, showing how the duty cycle of
an oscillator can be dynamically regulated by electrically coupled neurons.
2
In the third paper, Michail Zak abstracts away from the detailed properties of
neurons to study the collective properties of large networks. He links two different
concepts: a) the collective brain, as a set of simple units of processors (neurons) that
interact by exchanging signals without explicit global control, and where the
objective of each unit may be partially compatible or contradictory, Le., the units
may cooperate or compete; and b) a dynamical system, where a dynamical model is
represented by a system of ordinary differential equations with terminal attractors
and repellers, and it can be implemented only by analog elements. Zak takes
advantage of the mathematical tools that have been developed to analyze dynamical
systems, to propose a mathematical framework to study a dynamical system which
mimics collective purposeful activities of a set of units which process a "knowledge
base", in the form of a joint density function, without global control. In this system
global coordination is replaced by the probabilistic correlations between the units.
These correlations are learned during a long-term period of performing collective
tasks, and they are stored in the joint density function.
The understanding of the brain requires not only the explanation of phenomena
occurring at the level of neural networks dynamics, but also at another different
levels of complexity, which include overt animal behavior. In the fourth paper,
Michael A. Arbib presents a methodology he defines as "Schema Theory" to attempt
to bridge the gap between the different levels of analysis. This methodology offers
the possibility of developing studies under three different approaches: "Top-Down",
from behavior to neural mechanisms, "Bottom-Up", from neural mechanisms
towards explaining behavior, and "Middle-Out", bridging intermediate levels of
analysis. Although Arbib focuses on schema-theoretic models of sensorimotor
coordination phenomena in living beings, it is clear that the same tools can be used
to analyze similar phenomena in the design of complex automata (e.g., sensory
based robots). Thus, an important characteristic of Arbib's Schema Theory is that its
explanation of intelligent behavior in terms of the interactions among different
functional processes provides a bridge between Neuroscience and AI.
Finally, to complete this section's progression from membrane properties to the
interactions of more and more complex systems, Edmund Durfee offers a point of
view from Distributed Artificial Intelligence, showing how intelligent agents may
coordinate their activities. Such agents may be like schemas competing and
cooperating within the "head" of a single animal or robot, or may indeed be part of a
truly social organization bringing together many people (and their machines) to
collectively solve problems that are beyond their individual capabilities.