A practical cognition system is a large scaled complex system. Conventional method cannot design this kind of systems efficiently. To solve this problem, we proposed the individual evolutionary algorithm (IEA). The IEA is an efficient algorithm for obtaining networks with right sizes and high performance from given samples. Although it was proposed for the case of the nearest neighbor based MLP, it is useful for learning of more complex systems as well. In this research, we have extended the IEA in three directions: 1) On-line evolutionary learning, 2) Co-evolutionary learning, and 3) hierarchical evolutionary learning. On-line learning is necessary for evolutionary learning in changing environments. Co-evolutionary learning is useful for obtaining complex individuals from simple ones. And hierarchical evolutionary learning is the final stage for designing a complex system.