Consideration on singularities in learning theory by complex and real analysis

2016 Fiscal Year Final Research Report

• Project/Area Number: 23540224
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Basic analysis
• Research Institution: Nihon University
• Principal Investigator: AOYAGI Miki 日本大学, 理工学部, 准教授 (90338434)
• Project Period (FY): 2011-04-28 – 2017-03-31
• Keywords: 学習係数 / 特異点解消 / log canonical threshold

Outline of Final Research Achievements

The purpose of a learning system is to estimate an unknown true density function that generates the data. Recently, the WAIC model selection method, which is a generalized AIC, has been considered for estimating a probability function from data in a singular learning system. The learning coefficient in Bayesian estimation serves to measure the learning efficiency, and has an important role in the WAIC method. Mathematically, the learning coefficient is the log canonical threshold of the relative entropy. In this study we consider learning coefficients and several theorems for our results.

Free Research Field

学習理論