Development and application of subspace learning methods to reveal data structure.

Nowadays, with the development of deep learning techniques, the capacity for accumulating information has been remarkably improved, leading to the production of more large-scale and complex datasets. As a result, unsupervised machine learning, which involves devising mathematical models based on the characteristics and structures of such big data, has gained increasing attention to explore what serves as the main interests for the whole society. We have been working on how to represent big data in a latent lower-dimensional vector space as compactly as possible, by carefully identifying the key information within high-dimensional data.