A multiagent data warehousing (MADWH) and multiagent data mining (MADM) approach to brain modeling and neurofuzzy control

Based on the hypothesis that the brain is a society of semiautonomous neural agents and full autonomy is the result of coordination of semiautonomous functionalities, a multiagent data warehousing (MADWH) and multiagent data mining (MADM) approach is presented for brain modeling and illustrated with robot motion control. An algorithm named Neighbor-Miner is proposed for MADWH and MADM. The algorithm is defined in an evolving dynamic environment with semiautonomous neurofuzzy agents. Instead of mining frequent itemsets from customer transactions, the new algorithm discovers new neurofuzzy agents and mines agent associations in first-order logic for coordination that was once considered impossible in traditional data mining. While the Apriori algorithm uses frequency as a priori threshold, Neighbor-Miner uses agent similarity as a priori knowledge. The concept of agent similarity leads to the notions of agent cuboid, orthogonal MADWH, and MADM. Based on agent similarities and action similarities, Neighbor-Miner is presented and illustrated with brain modeling for robot control. The novelty of a multiagent data warehouse lies in its ability to systematically combine neurofuzzy systems, multiagent systems, database systems, machine learning, data mining, information theory, neuroscience, decision, cognition, and control all together into a modern multidimensional information system architecture * Corresponding author. Tel.: +1-912-486-7198; fax: +1-912-486-7672. E-mail address: [email protected] (W.-R. Zhang). 0020-0255/$ see front matter 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.ins.2003.05.011 110 W.-R. Zhang, L. Zhang / Information Sciences 167 (2004) 109–127 that is ideal for brain modeling of different animal species with manageable complexity. Although examples in robot control are used to illustrate the basic ideas, the new approach is generally suitable for data mining tasks where knowledge can be discovered collectively by a set of similar semiautonomous or autonomous agents from a geographically, geometrically, or timely distributed environment, especially in highdimensional scientific and engineering data environments. 2003 Elsevier Inc. All rights reserved.