Abstract
Most of neural networks are pattern classifier trained by an external supervisor. However, the real brain can acquire their behavior even without a supervisor. In this paper, a brief introduction of neural networks will be shown followed by the guiding principle of learning mechanism without a supervisor. First, the comparison between neural networks and electronic circuits will be given. Next, the network formation by backpropagation method will be introduced. The method is very effective, but it is a supervised learning, and it is very difficult to train the recurrent neural networks. Finally, a new unsupervised learning concept of Moderationism proposed by us will be presented, which is still under research but will be useful to understand the brain behaviors.
