栁原 宏和




School of Informatics and Data Science
Graduate School of Science, Probability and Mathematical Statistics Course

Challenge yourself to study theoretical approaches supporting data analysis.

Theoretical study for validating data analysis methods

My research relates to theoretical studies on statistics, including theoretical analysis for numerical validation and setting of new approaches of multivariable analysis, which is an approach for analyzing multivariable data with two or more variables. Recently, I am engaged in the development of methodologies for the analysis of multivariable data with a huge number of variables (high-dimension data) and an approach for theoretical validation of such methodologies. More concretely, the focal point of my studies is the development of variable selection model for determining the necessity of the relevant variable in the high-dimension multivariable data analysis method. My target is to develop a variable selection model in which the probability of the truly necessary variable being selected is 1, supposing that the number of samples is infinite (agreement).

Growing needs for high-dimension data analysis

Recently, the data amount which can be stored and analyzed is surging, so there is a growing need for high-dimension data analysis. Traditional multivariable analysis methods assess the theoretical validity based on asymptotic theory in which only the number of samples is infinite. Thus, such assessment is based on the condition that the number of samples is infinite. However, in some cases, traditional asymptotic theory does not work for high-dimension data. In some cases, even if a positive result is obtained by assessment based on asymptotic theory, such result turns out not to work for actual assessment based on definite samples. This problem can be avoided by the re-assessment by new asymptotic theory, namely, increasing the size of the dimension which is the number of variables in multivariable data along with the number of samples. It is a very pleasant moment when a methodology which ought to work well under the traditional assessment turns out to be the other way around under the new assessment.

Gravity of theoretical study

Data science study requires the knowledge and competency of taking a wide view of data while paying attention to the background of data. A student who is familiar with and likes mathematical expressions would have good aptitude. Nowadays, less emphasis is placed on theoretical studies, however, it is impossible to use any analytical method proactively without theoretical validity. For this reason, I find theoretical studies to be very important. My goal is to spread statistics-based data analysis method throughout society. Why not join me for this theoretical study?