Co-evolutionary Multi-agent Learning by means of Genetic Programming

点击翻译按钮获取中文摘要

It is not easy to describe in advance the program which controls more than one agent effectively. This is due to the difficulty with multi-agent programming for the communication, cooperation, synchronization and conflict resolution. Thus, we propose a multi-agent learning scheme based on the co-evolutionary method for the purpose of the automatic generation of the programs for many agents. Our approach is based upon the evolutionary computation, i.e., genetic programming and genetic algorithms.The established multi-agent learning system has been applied to several practical problems, and the validity has been shown. The application areas include the wider field of examples as described below :(1) Adaptive WWW search by the Web robot.(2) Automatic synthesis of analog circuits, such as an asymmetric bandpass filter.(3) Inference of gene regulatory networks, i.e., Weaver models or differential equation systems, for genome informatics.(4) Interactive music composition.(5) Evolutionary quantum circuit design, such as quantum teleportation and oracle problems.(6) Financial data prediction, such as Nikke225 or foreign exchange rate.