講演抄録/キーワード |
講演名 |
2010-11-04 15:00
[ポスター講演]ブレグマン距離を用いた密度比推定の統一的枠組み ○杉山 将(東工大)・鈴木大慈(東大)・金森敬文(名大) IBISML2010-64 |
抄録 |
(和) |
(事前公開アブストラクト) Estimation of the ratio of probability densities has attracted a great deal of attention
since it can be used for addressing various statistical paradigms
such as non-stationarity adaptation, two-sample test,
outlier detection, mutual information estimation,
dimensionality reduction, independent component analysis, causal inference,
conditional density estimation, and probabilistic classification.
A naive approach to density ratio approximation
is to first estimates numerator and denominator densities separately and
then take their ratio.
However, this two-step approach does not perform well in practice,
and methods for directly estimating the density ratio without going
through density estimation have been explored,
including methods based on
moment matching,
probabilistic classification,
density matching,
and density-ratio fitting.
The contributions of this paper are three folds:
First, we give a comprehensive review
of existing density ratio estimation methods
and discuss their pros and cons.
The second contribution is that we propose a new framework of density ratio estimation
in which a density-ratio model is fitted to the true density-ratio under the Bregman divergence.
Our new framework includes all the above existing approaches as special cases,
and is substantially more general.
Thus, it provides a unified view of various density ratio estimation methods.
Finally, we develop a robust density ratio estimation method
under the power divergence, which is a novel instance in our framework. |
(英) |
Estimation of the ratio of probability densities has attracted a great deal of attention
since it can be used for addressing various statistical paradigms
such as non-stationarity adaptation, two-sample test,
outlier detection, mutual information estimation,
dimensionality reduction, independent component analysis, causal inference,
conditional density estimation, and probabilistic classification.
A naive approach to density ratio approximation
is to first estimates numerator and denominator densities separately and
then take their ratio.
However, this two-step approach does not perform well in practice,
and methods for directly estimating the density ratio without going
through density estimation have been explored,
including methods based on
moment matching,
probabilistic classification,
density matching,
and density-ratio fitting.
The contributions of this paper are three folds:
First, we give a comprehensive review
of existing density ratio estimation methods
and discuss their pros and cons.
The second contribution is that we propose a new framework of density ratio estimation
in which a density-ratio model is fitted to the true density-ratio under the Bregman divergence.
Our new framework includes all the above existing approaches as special cases,
and is substantially more general.
Thus, it provides a unified view of various density ratio estimation methods.
Finally, we develop a robust density ratio estimation method
under the power divergence, which is a novel instance in our framework. |
キーワード |
(和) |
/ / / / / / / |
(英) |
density ratio / Bregman divergence / logistic regression / kernel mean matching / Kullback-Leibler importance estimation procedure / least-squares importance fitting / / |
文献情報 |
信学技報, vol. 110, no. 265, IBISML2010-64, pp. 33-44, 2010年11月. |
資料番号 |
IBISML2010-64 |
発行日 |
2010-10-28 (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 |
IBISML2010-64 |
|