(Return to ICNN97 Homepage) (Return to ICNN'97 Agenda)



ICNN'97 FINAL ABSTRACTS


CS: COGNITIVE SCIENCE AND COGNITIVE NEUROSCIENCE


ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS1A Paper Number: 423 Oral

Neural network model for spatial memory

Kunihiko Fukushima and Yoshio Yamaguchi

Keywords: spatial memory correlation matrix memory imagery layer

Abstract:

This paper offers a neural network model that can memorize and recall spatial maps. This is an improved version of the model we proposed previously. When driving through a place we have been before, we can recall and imagine the scenery that we cannot see yet but shall see soon. Triggered by the newly recalled image, we can also recall other scenery further ahead of us. The model emulates such a recalling process.

In the computer simulation, we prepare a map of Europe and assume a situation where one (in this case, our model) makes a trip along railways. At first, the model wanders around Europe and memorizes fragmentary maps around it. After that, the model makes another trip along a neighboring rout. Whenever the model moves along a railway, the model recalls new maps ahead on the rout from the memory. Thus, an image covering a wide area is retrieved by a continuous chain process of recalling. Even though the traveling rout is novel to the model, the recalled maps are correct in most cases, if the model has made a trip in the neighborhood before.

When the model comes to a new place, it simply memorizes the map around it.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS1B Paper Number: 146 Oral

A role for the hippocampal system in the learning of distal cortical associations

J. G. Wallace and K. Bluff

Keywords: hippocampal system distal cortical associations Newel's model

Abstract:

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS1C Paper Number: 635 Oral

A neural network for decision making under the influence of reinforcement

Jean-Daniel Kant and Daniel S. Levine

Keywords: reinforcement decision making adaptive resonance theory

Abstract:

A neural network based on adaptive resonance theory, known as Categ_ ART, has previously been developed to model the actual process of human decision making and to discern the basis for the actual categorizations made, and applied to data on choices made among bank savings schemes. This network is further extended herein to include representations of the criteria for categorization decisions. The strength of a particular criterion representa- tion can be increased if that criterion successfully predicts the appropriate category, and decreased if it leads to ambiguity in the choice of category.

Moreover, the connections between criterion and category nodes can be modulated by selective attentional biases that may in turn be influenced by external reinforcement. Some possible analogies with frontal lobe function and with animal inductive learning results are discussed.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS1D Paper Number: 123 Oral

A neocortically-derived model of continuous contextual processing

Andreas Garzotto, Boris Aleksandrovsky, Gary Lynch and Richard Granger

Keywords: continuous contextual processing neocortical regions feedforward-feedback system

Abstract:

The architectural regularities shared among most neocortical regions suggest repeated functional units that confer core computational capabilities to otherwise very different cortical areas. This paper addresses the massive cortico-cortical feedforward-feedback system connecting most adjacent cortical areas and discusses a computational model derived from these feedforward-feedback loops. First results obtained using a partial implementation of the model show context-dependent pattern recognition capabilities such as generalization, noise tolerance, pattern completion, and cued associative recall, even with unsegmented input data.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS2A Paper Number: 185 Oral

Modelling the perceptual separation of concurrent vowels with a network of neural oscillators

Guy J. Brown and DeLiang Wang

Keywords: neural oscillators perceptual separation concurrent vowels

Abstract:

The ability of listeners to identify two simultaneously presented vowels is improved by introducing a difference in fundamental frequency between the vowels. We propose an explanation for this phenomenon in the form of a computational model of concurrent sound segregation, which is motivated by neurophysiological evidence of oscillatory firing activity in the higher auditory system. In the model, the perceptual grouping of auditory peripheral channels is coded by synchronised oscillations in a neural oscillator network.

Computer simulations confirm that the model qualitatively matches the double vowel identification performance of human listeners.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS2B Paper Number: 307 Oral

Multiresolution elementary tonotopic features for speech perception

Elaine Y. L. Tsiang

Keywords: elementary tonotopic features speech perception fixed-weight FIR neural network

Abstract:

We define multiresolution elementary tonotopic features (ETFs) in general, and present specific functions and decompositions for computing them. Such decomposition, when cast in the form of local, fixed-weight FIR neural networks, have definite architectures. Results of their use as front-end inputs to a speaker-independent continuous-speech phoneme recognizer are encouraging. We analyze the dependence of the recognition performance on the various ETFs at different levels of resolution.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS2C Paper Number: 533 Oral

A minimal neural network for target searching and danger avoidance by smell

Wee Kheng Leow and Hsueh Chyi Yee

Keywords: minimal neural network target searching danger avoidance

Abstract:

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS2D Paper Number: 549 Oral

Hierarchical classification of odor quality based on dynamical property of neural network and olfactory cortex

Tetsuya Oyamada, Yoshiki Kashimori and Takeshi Kambara

Keywords: Hierarchical classification odor quality dynamical property of neural network olfactory cortex

Abstract:

We study a mechanism of odor classification in olfactory cortex based on the hypothesis that components of an odor and their mixing ratio are encoded into a temporal sequence of spatial activity patterns, that is , a limit cycle attractor in olfactory bulb. We present a functional network model of olfactory cortex which consists of three, rostral, middle, and caudal, compartments.

When a temporal sequence of spatial firing patterns is injected from the olfactory bulb to the network model, neural activity states of rostral, middle and caudal compartments are fixed to the spatial patterns corresponding to the strong , middle and weak components, respectively. Each compartment recognizes a relevant odorant component.

An odor quality is recognized based on the combination of three fixed patterns. The stronger the mixing ratio of a component is , the earlier the component is recognized. This is a hierarchical classification.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS2E Paper Number: 523 Oral

Neural field model of oculomotor preparation and disengagement

T. Trappenberg, S. Simpson, R. M. Klein, D. P. Munoz, M. C. Dorris and P. McMullen

Keywords: Neural field model oculomotor preparation and disengagement gap effect

Abstract:

Seeking to develop an artificial neural network model that can match the behavior of the primate oculomotor system and is solidly grounded in the most up-to-date neuroscientific knowledge, we chose, from among extant models, an abstract, neural field model (Kopecz's) as a starting point. Originally designed to reproduce the ``gap effect'', this model appealed to us because: 1) its motor programming units appear analogous to ``buildup'' neurons in the middle layer of the superior colliculus, and 2) in contrast to most other models, this one explicitly allows for intentional inputs to motor programming. Our behavioral data reveals a small effect of voluntary preparation (target location probability) in the overlap and step conditions which becomes very large in the gap condition (fixation is removed 200 ms prior to the appearance of the target). Although the original model could not match this pattern, a match was achieved by modifying the temporal pattern of intentional inputs, a modification that also produced a better match between the dynamic properties of our ``abstract'' motor programming units and actual ``buildup'' neurons.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CS2F Paper Number: 91 Oral

World-centered representation for neural networks

Lei Guo

Keywords: pattern recognition invariance full-parallel computation

Abstract:

A basic problem for pattern recognition is invariance. However, there have been a large amount of facts to make known that a neural network itself seems difficult to do so because no neural models have really revealed the neural principle of the invariant recognition. This article argues that the traditional full-parallel computation of neural networks that has excluded many useful series computation methods seems improper for the invariance. We introduce a series search mechanism into neural computing. The world-centered recognition is acquired by a pattern-centered memory plus a space search.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CSP2 Paper Number: 76 Poster

Natural properties in an artificial neural network

Stefan C. Kremer

Keywords: Natural properties artificial neural network feature identification

Abstract:

This paper describes a simple artificial neural network and a novel training algorithm for feature identification in the context of stereo vision. Despite many differences in the underlying operational principles of biological vision systems and the artificial neural network described, the two approaches exhibit surprisingly similar behavior. This gives confidence in the use of artificial neural networks to model cognitive tasks.

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CSP2 Paper Number: 114 Poster

Visual position and orientation orders

Zhiyong Yang and Songde Ma

Keywords: Visual position orientation regularization theory

Abstract:

_____

ICNN97 Cognitive Science & Cognitive Neuroscience Session: CSP2 Paper Number: 113 Poster

Phase transitions and bifurcation in visual perception

Zhiyong Yang and Songde Ma

Keywords: Phase transitions bifurcation visual perception


Web Site Author: Mary Lou Padgett (m.padgett@ieee.org)
URL: http://www.mindspring.com/~pci-inc/ICNN97/papercs.htm
(Last Modified: 30-Apr-1997)