Motoaki Kawanabe
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Motoaki Kawanabe has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
Masashi Sugiyama
9 shared publications- Masashi Sugiyama
RIKEN Center for Advanced Intelligence Project
3 shared publications - Klaus-Robert Müller
Technische Universität Berlin
2 shared publications - Paul von Bünau
Technical University of Berlin Franklinstr. 28/29, 10587 Berlin, Germany
2 shared publications - Taiji Suzuki
University of Tokyo
2 shared publications - Alexander Binder
ISTD Pillar, Singapore University of Technology and Design, Singapore
1 shared publication
Latest Publications
- 30 March 2012
- Published by MIT Press
- p. 240-242
Conclusions and Future Prospects
This chapter summarizes the main themes covered in the preceding discussions and discusses future prospects. This book has provided a comprehensive overview of theory, algorithms, and applications of machine learning under covariate shift. Beyond covariate shift adaptation, it has been shown recently that the ratio of probability densities can be used for solving machine learning tasks. This novel machine learning framework includes multitask learning, privacy-preserving data mining, outlier detection, change detection in time series, two-sample test, conditional density estimation, and probabilistic classification. Furthermore, mutual information—which plays a central role in information theory—can be estimated via density ratio estimation.
- 30 March 2012
- Published by MIT Press
- p. 240-242
Conclusions and Future Prospects
This chapter summarizes the main themes covered in the preceding discussions and discusses future prospects. This book has provided a comprehensive overview of theory, algorithms, and applications of machine learning under covariate shift. Beyond covariate shift adaptation, it has been shown recently that the ratio of probability densities can be used for solving machine learning tasks. This novel machine learning framework includes multitask learning, privacy-preserving data mining, outlier detection, change detection in time series, two-sample test, conditional density estimation, and probabilistic classification. Furthermore, mutual information—which plays a central role in information theory—can be estimated via density ratio estimation.
- 30 March 2012
- Published by MIT Press
- p. 225-237
Applications of Active Learning
This chapter describes real-world applications of active learning techniques: sampling policy design in reinforcement learning and wafer alignment in semiconductor exposure apparatus.
- 30 March 2012
- Published by MIT Press
- p. 225-237
Applications of Active Learning
This chapter describes real-world applications of active learning techniques: sampling policy design in reinforcement learning and wafer alignment in semiconductor exposure apparatus.
- 30 March 2012
- Published by MIT Press
- p. 215-224
Active Learning with Model Selection
This chapter examines the problem of active learning with model selection. Model selection and active learning are two important challenges for successful learning. A natural desire is to perform model selection and active learning at the same time, that is, we want to choose the best model and the best training input points. However, this is actually a chicken-and-egg problem since training input samples should have been fixed for performing model selection and models should have been fixed for performing active learning. The chapter discusses several compromise approaches, such as the sequential approach, the batch approach, and the ensemble approach. Using numerical examples, limitations of the sequential and batch approaches are pointed out, and the usefulness of the ensemble active learning approach is demonstrated.
- 30 March 2012
- Published by MIT Press
- p. 215-224
Active Learning with Model Selection
This chapter examines the problem of active learning with model selection. Model selection and active learning are two important challenges for successful learning. A natural desire is to perform model selection and active learning at the same time, that is, we want to choose the best model and the best training input points. However, this is actually a chicken-and-egg problem since training input samples should have been fixed for performing model selection and models should have been fixed for performing active learning. The chapter discusses several compromise approaches, such as the sequential approach, the batch approach, and the ensemble approach. Using numerical examples, limitations of the sequential and batch approaches are pointed out, and the usefulness of the ensemble active learning approach is demonstrated.
- 30 March 2012
- Published by MIT Press
- p. 182-214
Active Learning
This chapter examines the problem of active learning. The goal of active learning is to find the most “informative” training input points so that learning can be successfully achieved from only a small number of training samples. In the active learning scenario, covariate shift—mismatch of training and test input distributions—occurs naturally occurs since the training input distribution is designed by users, while the test input distribution is determined by the environment. Thus, covariate shift is inevitable in active learning. The chapter introduces active learning methods for regression in light of covariate shift. Their mutual relation and numerical examples are also shown. Furthermore, these active learning methods are extended to the pool-based scenarios, where a set of input-only samples is provided in advance and users want to specify good input-only samples to gather output values.
- 30 March 2012
- Published by MIT Press
- p. 182-214
Active Learning
This chapter examines the problem of active learning. The goal of active learning is to find the most “informative” training input points so that learning can be successfully achieved from only a small number of training samples. In the active learning scenario, covariate shift—mismatch of training and test input distributions—occurs naturally occurs since the training input distribution is designed by users, while the test input distribution is determined by the environment. Thus, covariate shift is inevitable in active learning. The chapter introduces active learning methods for regression in light of covariate shift. Their mutual relation and numerical examples are also shown. Furthermore, these active learning methods are extended to the pool-based scenarios, where a set of input-only samples is provided in advance and users want to specify good input-only samples to gather output values.
- 30 March 2012
- Published by MIT Press
- p. 124-136
Relation to Sample Selection Bias
- 30 March 2012
- Published by MIT Press
- p. 103-123
Direct Density-Ratio Estimation with Dimensionality Reduction
This chapter discusses a dimensionality reduction scheme for density-ratio estimation, called direct density-ratio estimation with dimensionality reduction (D3; pronounced as “D-cube”). The basic idea of D3 is to find a low-dimensional subspace in which training and test densities are significantly different, and estimate the density ratio only in this subspace. A supervised dimensionality reduction technique called local Fisher discriminant analysis (LFDA) is employed for identifying such a subspace. The usefulness of the D3 approach is illustrated through numerical experiments.
- 30 March 2012
- Published by MIT Press
- p. 103-123
Direct Density-Ratio Estimation with Dimensionality Reduction
This chapter discusses a dimensionality reduction scheme for density-ratio estimation, called direct density-ratio estimation with dimensionality reduction (D3; pronounced as “D-cube”). The basic idea of D3 is to find a low-dimensional subspace in which training and test densities are significantly different, and estimate the density ratio only in this subspace. A supervised dimensionality reduction technique called local Fisher discriminant analysis (LFDA) is employed for identifying such a subspace. The usefulness of the D3 approach is illustrated through numerical experiments.
- 30 March 2012
- Published by MIT Press
- p. 72-102
Importance Estimation
- 30 March 2012
- Published by MIT Press
- p. 20-45
Function Approximation
- 1 March 2011
- Published by Elsevier in Neural Networks
- Vol. 24 (2) , 183-198
Direct density-ratio estimation with dimensionality reduction via least-squares hetero-distributional subspace search
- 1 March 2011
- Published by Elsevier in Neural Networks
- Vol. 24 (2) , 183-198
Direct density-ratio estimation with dimensionality reduction via least-squares hetero-distributional subspace search