The Philosophical Exploration Of Machine Learning

The philosophical exploration of Machine Learning mainly expounds the foundamental issues concerning machine learning, of which contains the standpoints of machine learning, the problem of induction, the explicating of probability, learnable question, and knowledge discovery etc.The research frontier of machine learning comes into being the overview of machine learning. This dissertation selects Machine Learning and Journal of Machine Learning, according to the principle of scientometrics and knowledge domain mapping, applying the software of CiteSpacell, the paper tailors the visualisational pattern of machine learning.The research fronts reveal the nine knowledge groups, of which include reinforcement learning, classification, computation learing theory and so on. The evolutionary path of machine learning is that the reinforcement learning is a relatively independent research steam, based on the classification technique, the data mining has two bracnches:One is the improvement of classification algorithms, the other is computation learning theory.The possible fulfilling pathways of machine learning cover various fields, such as the problem of induction, probility and statistic, and learnable problem etc. The ontology of machine learning is that existing a universal machine that have adequate logical functions can simulate all of the logical rules.There have two ways to implement the machine learning:one is the model of deductive inference, the other is the model of induction and statistics. In the fisrt method, the hypothesis was input into the computer together with empirical data, through certain inference, the machine could get new knowledge or improve the system's performance; while in the second, the combination of probility and statistics, adding to the belief of simplicity, which make it possible to confirm a hypothesis from the Hypothesis-Space."The logic of discovery" is a significant thesis of philosophy of science. Knowledge discovery is performed by the algorithms of machine learning. The nature of these leaning algorithms is the combination of recursions, iterations and other factor, the algorithms of machine learning implicate that the knowledge discovery is a nonlinear process, the knowledge discovery is an approximatively periodic process, the strange attractor can united the competitive paradigm and the minimal difference can result great diverstiy. Therefor the knowledge system has fractal structure and strange attractor. So there are some rules about the process of knowledge discovery.Approaching the end of this dissertation, I draw a framework of philosophy of machine learning, and discuss the relationships between science and philosophy of machine learning.