@techreport{oai:ipsj.ixsq.nii.ac.jp:00204113,
 author = {西山, 育宏 and 渡辺, 秀行 and 片桐, 滋 and 大崎, 美穂 and Ikuhiro, Nishiyama and Hideyuki, Watanabe and Shigeru, Katagiri and Miho, Ohsaki},
 issue = {46},
 month = {Mar},
 note = {大幾何マージン最小分類誤り学習法は，標本空間内に学習標本を仮想的に増加させ，最小分類誤り確率，即ちベイズ誤りの推定精度を向上させる効果を持つことが示唆されている．しかし，その効果に関する従来の評価は，多岐にわたる推定精度影響要因を必ずしも十分に制御せずに行われていた．本研究では，分類器のクラス境界表現力などの制御要因を網羅的に制御し，かつ交差検証法を用いた実験を通して，その効果の存在を明らかにする．, Previous studies suggested that the Large Geometric Margin-Minimum Classification Error (LGM-MCE) training method had the effect of virtually increasing training samples in a sample space and improving the quality of Bayes error (minimum classification error probability) estimation. However, those studies were simply conducted without sufficiently controlling various estimation-influencing factors. In this paper, we comprehensively control the influencing factors in the LGM-MCE training such as the capability of representing class boundaries (classifiers' model sizes) and clarify the existence of the effect through experiments using the cross-validation training/testing scheme.},
 title = {大幾何マージン最小分類誤り学習法のベイズ誤り推定力に関する実験的評価},
 year = {2020}
}