Research On Core Algorithms And Techniques In Brain Data Mining For Chinese Cognition

It's a real challenge for us to research brain, protect brain, develop brain, and create brain. With the development of brain science, a various kinds of new methods and techniques are applied. A mass of data is collected ceaselessly, which implicates much very important information. If the useful information could not be extracted and translated to intelligible knowledge from the data, it would lose the important signification. It is exigent to develop a method to analyze it availably. Brain data mining is becoming a necessary tools. The thesis focuses on the core techniques and algorithms in brain data mining for Chinese cognition, which is supported by MOST (2001CCA00700), MOE (KP0302) and Human Brain Project (HBP). The main topics are discussed in the thesis as follows:Analyzing particle swarm optimization (PSO) algorithm, its premature convergence is due to a decrease of velocity of particles in search space that leads to a total implosion and ultimately fitness stagnation of the swarm. A chaotic particle swarm optimization (CPSO) is introduced to overcome the problem of premature convergence. CPSO uses the properties of ergodicity, stochastic property, and regularity of chaos to lead particles' exploration. It supplies some power to the swarm system on sustainable development. Compared to PSO(s), Genetic Algorithms (GA) and Simulated Annealing (SA) on benchmark function optimization problems, the results show that CPSO prevents premature convergence effectively. It clearly outperforms than other considered approaches, especially for high dimension multi-modal optimization problems.A kind of knowledge relative reduction algorithm is proposed based on the discrete particle swarm optimization algorithm. The decision attribute support degree is applied in knowledge express system from rough set theory, and the support degree of the knowledge is supplied by condition attributes for the whole decision. So the relative importance degree and relative core are obtained by the discrete particle swarm searching, which overcomes the disadvantage of searchingthe minimum reduction from the discrirninability matrix. The tracing experimental results show that the approach is simple and effective in solving knowledge reduction. Compared the genetic algorithms, its performance is better with a higher success rate.An approach, Rough-set Neural-networks Data Mining (RNDM) is introduced, which integrates the techniques of artificial neural networks, rough set, and chaotic particle swarm optimization. Based on the rough set theory, attribute reduction is processed on the data under the consistent conditions. The neural network, constructed according to the reduced information, has a better topological structure with greatly reduced network scale while it keeps the good classification ability. Then neural network is used to study and predict data, at the same time to reduce the attributes under the inconsistent conditions. Chaotic particle swarm algorithm is applied to solve some optimization problems, especially for the related learning and reducing. Finally rule knowledge in the neural network is extracted by using rough set theory. It comes into the reduction and robustness ability for data mining. The empirical results clearly indicated that the approach can produce more effective and simpler rules quickly with good robustness.The thesis continues with a comprehensive look at reinforcing and extracting brain image's feature, and other processing involved brain data mining. An integrated system comes into being by brain data processing workflow, which consists of understanding the scope of the problem, understanding the data, preparing the brain data, and brain data mining. The remainder of the thesis is concerned with the application of the integrated techniques for brain data mining in Chinese cognition. This study demonstrates that it is feasible and efficient.