Unsupervised machine learning methods such as clustering and change detection are indispensable to various real-world data processing tasks. However, due to its vague formulation, studies of unsupervised learning tend to be ad-hoc, and thus development of unsupervised learning methods is still far behind supervised learning. The project aims at overcoming this difficulty by providing a systematic approach to unsupervised learning based on information measures. The PI and his group developed various information-based machine learning algorithms, including clustering, independence testing, object matching class-imbalance adaptation, change detection, and canonical dependency analysis. Furthermore, they explored fundamental data processing paradigms for further improving accuracy and robustness of information estimators in high dimensional problems. Through this project, they advanced the field of unsupervised learning by providing a novel systematic approach based on information measures.