Before Deep Learning’s successful recognitions, it was required for human beings to design characteristics to be recognized, but it is now possible for Deep Learning to learn structure of the data without such design. Multiple neural network learns at each layer and simple characteristics learned at the layer close to input can be used for other learning and recognition. For example, Deep Learning network for image recognition creates network detecting simple edges at data input level, and this network is effective in font recognition as well as general object recognition.