Adaptive Behavior, 8 (3/4)

We present a modified version of Schmajuk and Thieme's (1992) neural network model of spatial navigation. The new model differs from the original in several ways. First, whereas the early model assumed no a priori knowledge of the space to be explored, the present model assumes a representation of the environment as a set of potentially connected locations. Second, whereas in the original model the decision as to what place to move to next is based on the comparison of the predictions of the goal when each of the alternative places is briefly entered; in the present paper this decision is based on the comparison of the activation of each of the alternative places when the goal is activated. Computer simulations show that the present network offers a novel description of latent learning in terms of the competition between exploration and exploitation.

Key Words

exploration; neural networks; cognitive map

Pages 225-254

Learning Plans without a priori Knowledge

By Ron Sun and Chad Sessions

Abstract

This paper is concerned with the autonomous learning of plans in probabilistic domains without a priori domain-specific knowledge. In contrast to existing reinforcement learning algorithms that generate only reactive plans, and existing probabilistic planning algorithms that require a substantial amount of a priori knowledge in order to plan, a two-stage bottom-up process is devised in which first reinforcement learning/dynamic programming is applied, without the use of a priori domain-specific knowledge, to acquire a reactive plan, and then explicit plans are extracted from the reactive plan. Several options for plan extraction are examined, each of which is based on a beam search that performs temporal projection in a restricted fashion, guided by the value functions resulting from reinforcement learning/dynamic programming. Some completeness and soundness results are given. Examples in several domains are discussed that together demonstrate the working of the proposed model.

Pages 255-266

Effects of Genetic Dominance on Runaway Sexual Selection

By Mohamed A. F. Noor

Abstract

Distinguishing among theories of sexual selection requires that one develop diagnostic predictions that can be tested in living systems. Recently, genetic studies of female species preferences in Drosophila supported the predictions of a model of sexual selection through pleiotropy with adaptive traits: preferences generally behaved as recessive characters. However, the dominance predictions of female preferences resulting from runaway sexual selection have not been investigated. Here, I present an extension of previous simulation models of runaway sexual selection by varying the dominance of the female preference and incorporating genetic drift. I show that runaway sexual selection is generally more likely to favor the evolution of dominant female preferences than recessive ones. Also, in contrast to the results of a previous study, dominant preferred male characters spread more quickly by runaway sexual selection than recessive ones under some conditions. Overall, the predictions derived from this model of runaway sexual selection are not supported by empirical data on the genetic basis of species preferences, suggesting that runaway sexual selection may not be a major force in the evolution of such preferences. More empirical studies will be necessary to further evaluate both the predictions and the assumptions of this model, however.

Key Words

runaway sexual selection; sexy son; diploid model; dominance

Pages 267-296

Studies of Neuromodulation of Oscillatory Systems in Aplysia, by Means of Genetic Algorithms

By Dilip Deodhar and Irving Kupfermann

Abstract

Neural modeling methods were used to obtain insights into the role of neuromodulatory cotransmitters. Although the work was guided by specific experimental observations of feeding in Aplysia, it was meant as a more general treatment of behavioral systems. Genetic algorithms were used to evolve the parameters needed to permit a simple two-neuron circuit to oscillate and contract muscles that execute rhythmic feeding responses. The evolved circuits were found to possess a number of ``emergent'' properties not specifically selected for. The fitness of the circuit decreased under a variety of conditions, particularly when the rate of the rhythmic program increased. The fitness of the system could be restored when an autoregulatory cotransmitter system was added and served to dynamically alter the parameters of the muscles that generated the behavior.

Key Words

oscillators; cotransmitters; modulation; central pattern generators

Pages 297-312

Evolution and Learning in Neural Networks: Dynamic Correlation, Relearning and Thresholding

By Brian Carse and Johan Oreland

Abstract

This contribution revisits an earlier discovered observation that the average performance of a population of neural networks that are evolved to solve one task is improved by lifetime learning on a different task. Two extant, and very different, explanations of this phenomenon are examined-dynamic correlation, and relearning. Experimental results are presented which suggest that neither of these hypotheses can fully explain the phenomenon. A new explanation of the effect is proposed and empirically justified. This explanation is based on the fact that in these, and many other related studies, real-valued neural network outputs are thresholded to provide discrete actions. The effect of such thresholding produces a particular type of fitness landscape in which lifetime learning can reduce the deleterious effects of mutation, and therefore increase mean population fitness.

Key Words

genetic algorithm; machine learning; neural networks

Pages 129-136

The Mathmatics of Nepotism

By Jason Noble

Review of Foundations of Social Evolution, Steven A. Frank. Princeton University Press, 1998.

back to TOC, back to top

Adaptive Behavior, 8 (3/4)

Adaptive Behavior

Volume 8, Number 3-4

Winter 2000

Table of Contents

Horatiu Voicu and Nestor Schmajuk

Exploration, Navigation and Cognitive Mapping