Learning Under Distribution Mismatch

Independent and identically and distribution (i.i.d) assumption is a basis for classical learning methods, where the test data instances follow the same probability distribution as the training instances. However, the data evolve over time and change from one domain to another, thus the training data may be outdated and not enough representatives for the distribution of the test data. Transfer learning and domain adaptation, which is a general subfield of machine learning, aims to identify, extract and transfer the relevant useful knowledge from one or more source domain/task for learning in a target domain/task. This allows the domains, tasks, and distributions of training and test data to be different. There are three main issues on transfer learning: what to transfer; how to transfer; when to transfer. This book focuses on domain adaptation methods, aiming to develop effective solutions for real-world problems, especially biomedical healthcare systems. It contains two major parts: 1. To develop a new method for domain adaptation, in order to improve the efficiency, performance, and usability of real-world 2. To develop and analyze a general framework for automatic sleep staging.