講演抄録/キーワード |
講演名 |
2014-11-17 17:00
[ポスター講演]Target Shift Adaptation for Conditional Density Estimation and Regression ○Tuan Duong Nguyen(Tokyo Inst. of Tech.)・Marthinus Christoffel du Plessis・Masashi Sugiyama(Univ. of Tokyo) IBISML2014-55 |
抄録 |
(和) |
(事前公開アブストラクト) We consider the problem of domain adaptation under the target shift scenario, where the target marginal distribution p(y) changes between the training and testing phases, while the target-conditioned covariate distribution p(x|y) remains unchanged. Although various methods for mitigating target shift in classification (a.k.a. class-prior change) have been developed so far, few methods can be applied to continuous targets. In this paper, we propose methods of continuous target shift adaptation in regression and conditional density estimation. More specifically, our contribution is a novel importance weight estimator for continuous targets. Through experiments, the usefulness of the proposed methods is demonstrated. |
(英) |
We consider the problem of domain adaptation under target shift scenario, where the target marginal distributions p(y) differ between the training and testing phases, but the target-conditional covariate distribution p(x|y) remains unchanged. Although various methods for mitigating target shift in classification (a.k.a. class-prior change) have been developed so far, few methods can be applied to continuous targets. We propose methods for continuous target shift adaptation in supervised learning tasks. More specifically, our contribution is a novel importance weight estimator for continuous targets. Through experiments, the usefulness of the proposed methods is demonstrated. |
キーワード |
(和) |
/ / / / / / / |
(英) |
target shift adaptation / importance weight / conditional density estimation / regression / L2-distance / / / |
文献情報 |
信学技報, vol. 114, no. 306, IBISML2014-55, pp. 155-160, 2014年11月. |
資料番号 |
IBISML2014-55 |
発行日 |
2014-11-10 (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 |
IBISML2014-55 |