Page 127 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

116 CHAPTER 6 Evolving and Spiking Connectionist Systems

into larger neural network structures, tightly coupling learning and fuzzy reasoning
rules into connectionist structures, was initiated by Professor Takeshi Yamakawa
and other Japanese scientists [24]. Many models of fuzzy neural networks are
developed based on these principles [22,25,26].

1.5 EVOLUTIONARY COMPUTATION (EC): LEARNING PARAMETER
VALUES OF ANN THROUGH EVOLUTION OF INDIVIDUAL MODELS
AS PART OF POPULATIONS OVER GENERATIONS
Species learn to adapt through evolution (e.g., crossover and mutation of genes) in
populations over generations, always improving their features to survive in nature.
This principle, pioneered by Charles Darwin, called natural evolution [27,28],
was used to optimize parameters of ANN and AI systems for a better performance
and to select the optimum ANN and AI model among many. A schematic diagram of
an EC process is given in Fig. 6.4.
EC has borrowed principles from natural evolution, such as:

• Genes are carrier of information that realize both stability and plasticity;
• A chromosome defines an individual;
• Individuals create a new generation of individuals through crossover;
• Only the fittest individuals survive in a population to create the next generation;
• Selection criteria are based on ranking principles;
• Mutation is applied when there is no improvement of individuals over
generations.
EC is used as part of AI and ANN creation as a population/generation-based opti-
mization method to optimize the parameters of an ANN (e.g., connection weights,
number of neurons, learning rate, etc.), in order to select the best performing ANN
model (e.g., higher classification accuracy; minimum predicting error) [29,30].

FIGURE 6.4
A schematic diagram of an evolutionary computation process [22].