講演抄録/キーワード |
講演名 |
2010-09-05 10:40
Improving the Accuracy of Least-Squares Probabilistic Classifiers ○Makoto Yamada・Masashi Sugiyama(Tokyo Inst. of Tech.)・Gordon Wichern(MIT Lincoln Lab)・Jaak Simm(Tokyo Inst. of Tech.) PRMU2010-60 IBISML2010-32 |
抄録 |
(和) |
(事前公開アブストラクト) Least-Squares Probabilistic Classifier (LSPC) has been demonstrated to be a computationally-efficient and accurate classification method. However, since LSPC heuristically rounds up its negative parameters to zero as post-processing, its performance can be degraded if many parameters take negative values. In this paper, we propose not to round up negative parameters of the classifier, but the classifier's negative outputs. Through extensive experiments on benchmark and real-world image classification tasks, we show that the proposed post-processing method tends to improve the classification accuracy, while the computational advantage of the original LSPC is kept unchanged. |
(英) |
Least-Squares Probabilistic Classifier (LSPC) has been demonstrated to be a computationally-efficient and accurate classification method. However, since LSPC involves a post-processing step of rounding up its negative parameters to zero for assuring learned probabilities to be non-negative, its classification performance can be unexpectedly changed. In order to avoid this problem, we propose a simple alternative scheme that directly rounds up the classifier's negative outputs, not negative parameters. Through extensive experiments including real-world image classification and audio tagging tasks, we demonstrate that the proposed modification significantly improves the classification accuracy, while the computational advantage of the original LSPC is kept unchanged. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Least-Squares Probabilistic Classifier / Kernel Logistic Regression / Density Ratio / PASCAL VOC 2010 / Freesound / / / |
文献情報 |
信学技報, vol. 110, no. 188, IBISML2010-32, pp. 45-50, 2010年9月. |
資料番号 |
IBISML2010-32 |
発行日 |
2010-08-29 (PRMU, IBISML) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
PRMU2010-60 IBISML2010-32 |
|