The Computational Principles of Learning Ability

It has been quite a long time since artificial intelligence (AI) researchers in the field of computer science have stopped talking about simulating human intelligence or trying to explain how the brain works. Recently, represented by deep learning techniques, the field of machine learning has been experiencing unprecedented prosperity and some applications with near human-level performance bring researchers confidence to imply that their approaches are the promising candidates for understanding the mechanism of the human brain. However apart from several ancient philological criteria and some imaginary black box tests (Turing test, Chinese room) there is no computational explanation, definition or criteria about intelligence or any of its components. Based on the common sense that learning ability is one critical component of intelligence and from the viewpoint of mapping relations, this paper presents two laws which explain what ``learning ability" is, as we familiar with it and under what conditions a model can be acknowledged as a ``learning model". Furthermore, corresponding corollaries prove the existence of a common learning model (L), and by comparing with traditional learning theory with the theoretical framework proposed in this dissertation, the author explains why traditional classification models are not able to learn spontaneously.