Study Of Cognitive Neuroscience Mechanism Of Heuristic Problems Solving And Methods Of FMRI Data Analysis

Through several decades of development, artificial intelligence has made quite achievements, but also faces considerable bottlenecks. The ablities of problem-solving are the most general form of the human mind, and a concrete manifestation of the human high-level intelligence. The understanding and exploration of human brain problem-solving and information processing process can provide new ideas for artificial intelligence. With the purpose of"understanding brain, protecting brain, developing brain, creating brain", the study of cognitive neural mechanisms of problem-solving and information processing process has been studied in more and more different fields.This study is the main component of the National Natural Science Foundation project"The study of cognitive neuroscience about heuristic search in problem solving (ID: 60875075)", also is an exploratory research. Problem-solving information processing theory contends that problem solving is a process of operating different operators continuously to change the problem from the initial state to target state; the use of heuristic operator can improve the efficiency of problem-solving. Based on this theory, this study focuses on the cognitive neural mechanisms of the search and application of heuristic operator and information processing process during problem-solving process, and proposes the fMRI data analysis methods based on classification and clustering. Main tasks in this dissertation are as follows:(1) Based on the cognitive neural mechanisms of search and application of operator in problem solving, we proposed a new paradigm which is suitable to fMRI experiment, then design two cognitive experiments using this paradigm:â‘ the application of heuristic operators and retriving;â‘¡the seaching and selection of heuristic operators. The first cognitive experiment focuses on how human brain retrive and use a specific operator. The second experiment focuses on studying the neural mechanisms of human brain rational selection and searching for appropriate operators in problem state-space.(2) Multi-view analysis methods were used to explore the mechanism. The activated brain regions related to search and application of operator were explored through functional location analysis. On this basis, we used functional connectivity analysis to explore the connection patterns between the activated brain regions, and proposed the hypothesis of operator searching and the using cognitive processes. Then, these hypothesis were verified through a cognitive model established based on ACT-R. The degree of fitting with actual data and the ACT-R model is over 80%, and the good fitness implies that the model objectively reflects the possible information processing process, and the ACT-R model provides strong support for understanding neural mechanisms and information processing process of problem-solving.(3) Based on investigation of current fMRI classification method, a kind of data classification method, SVVC, has been proposed in this dissertation, researchers can use this mothod to explore the relationship between brain regions and certain cognitive process. We trained single voxel classifier and use AdaBoost algorithm to integrate classification results of single voxel classifier, and the accuracy of classification is high than 90%. This method can be used as a new ways of decoding high-level mind state based on fMRI. The study also compares classification accuracy of different feature selection methods and different classification algorithms. Experiments results show that the performance of using AdaBoost is the best, and the classification performance of selecting brain regions related to problem-solving is better than other regions.(4) This dissertation also proposed a fMRI data analysis method based on BOLD pattern clustering. We extract BOLD effects of all voxels of the whole brain, and use clustering algorithm to analyze these BOLD patterns, and find some typical BOLD patterns and their distribution within the brain. Experiments results show that there are some intrinsic BOLD patterns of the brain during the problems solving. On this basis, we analyz the cooperation working model of different brain regions through spatio-temporal analysis, the method could be an effective means of fMRI data analysis which can reveal the cooperation working pattern of the whole brain.In short, focusing on the cognitive neuroscience mechanism of heuristic operator searching and application while problem solving, this dissertation uses fMRI techniques and methodology of cognitive psychology, multi-methods, such as data mining and cognitive modeling, are all used to explore the neural mechanisms and information processing process. These cognitive neuroscience mechanism and information processing process of heuristic problem solving can provide some references and enlighten artificial intelligence hopely. The two new data-driven fMRI data analysis methods proposed in this dissertation can improve the level of fMRI data analysis.