Research Introduction


Intelligence computation:
    The Intelligence computation uses the enlightenment of biosphere rule, according to its principle, we imitate and design the algorithm of problem-solving . At present,it is make up of many content, such as: the technique of artificial neural network, genetic algorithm and evolutionary rules, simulated annealing technology and swarm intelligence technology, etc.
    Main Research achievements:
    1、According to the view of the evolution of the swarm intelligence , we propose that random ( connection) neural network model of the swarm intelligence .
    2、We use geometrical meaning of neurons to make learning problems into covering problem , giving the constructive machine learning methods - covering algorithm ,which is an important contribution to the statistical learning methods .
    3、According to the deficiency of the theory of SVM, we apply constructive method to the research of the theory SVM, main conclusions are drawn as follows:
        1)Prove the equivalence between the planning algorithm and the SVM algorithm
        2)Give the polynomial algorithm and geometry algorithm of solving the maximum edge
        3)Give the proof of the existence of kernel function
        4)Give the (polynomial) algorithm of solving the kernel function
        5)After researching the relations between support vectors' recognition with kernel function , we prove that a kernel function always exists in any nonempty subset D of sample sets, its support vectors set is just as equal to set D
        6)Give the constructive algorithm of kernel function
    4.We generalize the covering algorithm to kernel covering algorithm and introduce finite probability mixture model, we have made global optimization of algorithm and improved its precision and ability.

Quotient Space Theory:
    The core problem of artificial intelligence is to establish intelligent formalization model, according to the features of human intelligence, we put forward the quotient space granular computing theory. The given definition of fuzzy sets constructivity solved people's subjective criticisms of membership functions under a certain degree, and people found the nature fuzzy relation is the relationship between the sequence. We use the quotient attribute of quotient space to approximate original space attribute, finding quotient approximation and the second generation wavelet of multiresolution is equal. Quotient space theory is one of the three theories of the current international granular computing (the famous professor of Canada, Yao comments).