Kính mời quý Thầy Cô và các bạn sinh viên đến tham dự và trao đổi chuyên môn về nội dung Machine Learning của Giáo sư Hiroshi Mamitsuka (Trường Đại học Kyoto, http://www.bic.kyoto-u.ac.jp/pathway/mami/)
- Thời gian: từ 9g sáng ngày 17/11/2018 (Thứ bảy).
- Địa điểm: Phòng I42
Graph-regularized Data-integrative Machine Learning for Bioinformatics
Machine learning is now extensively used in science, engineering and business. Biology and medical sciences are also such an area with a variety of applications. A key feature of this field is various information with different data formats can be collected. A standard way of using this type of diverse data is to transform all given data into a format of one vector for each instance, i.e. eventually a matrix of examples and features. Then we can run an existing machine learning technique, such as deep learning for classification, over such a matrix. This manner, which ignores the original data formats, has serious drawbacks: 1) a lot of given information may be discarded, 2) a higher-dimensional data space may be made, etc. which makes this way far from a sophisticated manner from a machine learning viewpoint.
When we focus on two particular different types of inputs: vectors (or a matrix) and graphs, instead of transforming graphs into vectors, I will show you a more reasonable approach for integrating vectors with graphs under three general problem setting, classification (or regression), clustering and factorization. Our integration is to formulate each problem as a graph-regularized optimization problem. In fact these three settings are useful to solve three important and famous problems in bioinformatics: gene essentiality prediction, gene function prediction and predicting drug-target interaction, respectively. Finally I will point out the common or different properties in the above three formulations.
Hiroshi Mamitsuka worked in industry after he obtained the M.E. degree in information engineering from the University of Tokyo. At that time, he did research on machine learning and also data mining on a wide variety of applications, such as e-commerce campaign management, web access pattern mining, experimental design for drug discovery, etc. He got PhD in information sciences from the University of Tokyo, during he was working with a company, and then moved to academia, focusing more on methodological aspects of machine learning and biological applications. Currently he is a professor of Institute for Chemical Research, Kyoto University, and jointly appointed as a faculty of School of Pharmaceutical Sciences of the same university. Also he is a visiting professor of Aalto University in Finland.