Link Game Theory and Machine Learning in Theories and Applications
Tsai, Kuo Chun
MetadataShow full item record
Game theory is the study of mathematical models of strategic interaction among rational decision-makers. It has applications in all fields of social science, as well as in logic, systems science and computer science. Today, game theory applies to a wide range of behavioral relations. On the other hand, machine learning is the study of computer algorithms that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. In this dissertation, we try to advance machine learning with game theory in both applications and theories with the following three studies. We first link the machine learning with an existing theorem. Although game theory is a very profound study on distributed decision-making behavior and has been extensively developed by many scholars, many existing works rely on certain strict assumptions that might not be practical. With our proposed deep reinforcement learning model, we relaxed the assumptions in the theorem and study the opponent's behaviors instead of just learning the regions for corresponding strategies. We then study the potential of machine learning with the matching game in the mobile social networks edge caching problem. Low data transmission latency is one of the top priorities in the system. We proposed a machine learning model with long short-term memory structure to determine the relevance of the data to the user. Based on the machine-learned relevancy, we match the cached data with the base stations. With our proposed method, the data transmission latency within the network has highly reduced. We also bring machine learning technique into the oil and gas industry. The first arrival picking in performed to determine the underground structure. This process highly depends on the experts to manually look at the data which is too much time consuming and inconsistent judgment. With the help of machine learning with the semi-supervised learning and transform learning technique, we reduced the processing time and also produced consistent results by learning from different experts' experiences.