A Brain-inspired Neural Network Model Of Classical Conditioning And Its Applications

Implementing the ability of learning to enhance the self-adaption of robots with classical conditioning is the research hotspot of robots.Classical conditioning is the most fundamental learning system of animals,so it is very valuable to implement classical conditioning system on robots to enhance the ability of learning.This thesis discusses the theory model and implementation method of classical conditioning from the perspective of being inspired from brains.Spike timing-dependent plasticity is thought as the low level mechanism of classical conditioning.The traditional spike timing-dependent plasticity is discrete in time domain,which doesn't facilitate further analyzing and computing.This thesis proposes an equivalent continuous form of spike timing-dependent plasticity based on the research results from neural science,and eliminates the incompatibility of spike timing-dependent plasticity and classical conditioning in temporal scale.This thesis constructs a brain-inspired neural network model of classical conditioning which takes spike timing-dependent plasticity as its basic synaptic plasticity.In this model,environmental stimulus would change the activities of their corresponding neuron populations,and influence the synaptic efficiency.This model successfully explained and simulated the acquisition,extinction and blocking of classical conditioning.This thesis implements a classical conditioning system of robots.Based on this system,robots can learn the temporal relationship of stimulus of environment.Robots can predict the coming obstacle and make avoidance ahead.The model of classical conditioning this thesis proposes is well supported by the theory of neural science.This numeric computation and simulation experiments show this model can simulate the phenomenon of classical conditioning well.The implementation method of classical conditioning system this thesis proposes can endow the robots with the ability of establishing conditioned reflex,enhance the ability of learning of robots.