Cultural Learning In Evolutionary Spiking Neural Networks

During last few years we have witnessed a shift of the emphasis in the artificialneural networks community toward spiking neural networks. Motivated byneuroscience discoveries, many studies consider neural networks with spike-timing asan essential component in neural information processing. Theoretical studies haveshown that the computing power of spiking neural networks transferring time-domaininformation of a single neuron spike sequence is stronger than artificial neuralnetworks generally using Sigmoid as activation function. At present, the applicationof artificial neural networks in the field of population’s evolution and learning is verywide, considering the characteristics of spiking neural networks, this paper focuses onthe population’s evolution and learning behavior simulated by spiking neuralnetworks.Darwin’s natural selection theory shows that individuals with excellent traits willbe easier to get opportunity to reproduce and these excellent traits accumulating fromsuccessive generations become the major traits of the population during evolutionprocess. The evolution of spiking neural networks individual or autonomous agentdepends on the structure feature of spiking neural networks, this paper uses afeed-forward spiking neural networks model and designs some evolutionary operatorssuch as additive noise operator, weight replacement operator and columns exchangeoperator and so on. Under the action of evolutionary operators, we realize the weightsevolution of autonomous agent. Experimental results show that under the action ofevolutionary operators population’s adaptability to environment gradually increases.At the same time of population’s evolution, there exists learning behaviorbetween autonomous agents and this kind of learning behavior plays a key role toimprove the adaptability of the population. This paper studies the effects of culturallearning in spiking neural networks population, firstly, we choose teacher/pupil modelas the implementation scenario of cultural learning, and then we use a multi-spikeerror backpropagation learning algorithm based on gradient descent as theimplementation algorithm of cultural learning, finally we realize the cultural learningof spiking neural networks on the basis of evolutionary algorithm. In order to better show the important role of cultural learning on the evolutionary population, wecompare the fitness of evolutionary population and cultural learning population in ourexperiments and the results show that cultural learning can better improvepopulation’s adaptability to environment...