Machine Learning for Small Data

Abstract

The recent success of AI and machine learning has been based on the background that big data are available. On the other hand, as spreading application fields of machine learning, there is a growing need to apply machine learning to small data collected at a high cost. In this paper, we introduce machine learning frameworks such as sparse modeling and Bayesian modeling, which are suitable for small data. In addition, we introduce transfer learning for deep neural networks to deal with small data, and active learning and Bayesian optimization for efficient data acquisition. We discuss not only user aspects of machine learning but also some mathematical problems.