A structural representing and learning model based on biological neural mechanism

From the view of cognitive computational neuroscience, a direct representing method based on neural dynamics and graph theory is presented. Firstly, an assembly of neuron as well as its dynamics is defined. They directly represent the perceptual information of stimulus. Then a two layer neural network is designed to retain the features of that stimulus and generate the very neural circuit responding to it. This is achieved by the structure-learning algorithm. The circuit can also serve as an associative base whose credibility is decided by its connectivity. The direct representing method is of great significance in the research of semantic representation and semantic-driven inference in artificial intelligence as well as in artificial neural network researches. The exhibition of major physiological features of neural information processing distinguishes this model from the traditional ones.