New learning scheme for skill growing and structure expansion

One desirable feature of an intelligent system is the capability of expansion of the system structure as well as the learned techniques. In this study, we investigated the development of such a capability. The learning structure consists of a pool of neurons. For learning a primary skill, a genetic algorithm (GA) is adopted to search for weights for performing the skill. To develop more advanced techniques, a number of neurons are added. The GA is then used to search for weights within the new set of neurons and between the newly added neurons and the old ones. The original structure is not altered. This preserves the previously learned skills and a new skill can be established based on the existing ones. Although the purpose of this study is to resolve the problem of skill expansion upon completing a design, it is noted that the scheme can also be applied to relaxing the learning difficulty by decomposing a difficult skill. Attractive features of the new approach include its modular structure, system expansibility, requirement for less hardware resources and potentially faster learning