A comparative study of pairwise learning methods based on kernel ridge regression
- Author
- Michiel Stock (UGent) , Tapio Pahikkala, Antti Airola, Bernard De Baets (UGent) and Willem Waegeman (UGent)
- Organization
- Abstract
- Many machine learning problems can be formulated as predicting labels for a pair of objects. Problems of that kind are often referred to as pairwise learning, dyadic prediction, or network inference problems. During the past decade, kernel methods have played a dominant role in pairwise learning. They still obtain a state-of-the-art predictive performance, but a theoretical analysis of their behavior has been underexplored in the machine learning literature. In this work we review and unify kernel-based algorithms that are commonly used in different pairwise learning settings, ranging from matrix filtering to zero-shot learning. To this end, we focus on closed-form efficient instantiations of Kronecker kernel ridge regression. We show that independent task kernel ridge regression, two-step kernel ridge regression, and a linear matrix filter arise naturally as a special case of Kronecker kernel ridge regression, implying that all these methods implicitly minimize a squared loss. In addition, we analyze urdversality, consistency, and spectral filtering properties. Our theoretical results provide valuable insights into assessing the advantages and limitations of existing pairwise learning methods.
- Keywords
- DRUG-TARGET INTERACTIONS, SUPPORT VECTOR MACHINES, LEAST-SQUARES, REGULARIZATION ALGORITHMS, INTERACTION PREDICTION, MATRIX FACTORIZATION, NETWORKS
Downloads
-
(...).pdf
- full text
- |
- UGent only
- |
- |
- 434.04 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-8578969
- MLA
- Stock, Michiel, et al. “A Comparative Study of Pairwise Learning Methods Based on Kernel Ridge Regression.” NEURAL COMPUTATION, vol. 30, no. 8, 2018, pp. 2245–83, doi:10.1162/neco_a_01096.
- APA
- Stock, M., Pahikkala, T., Airola, A., De Baets, B., & Waegeman, W. (2018). A comparative study of pairwise learning methods based on kernel ridge regression. NEURAL COMPUTATION, 30(8), 2245–2283. https://doi.org/10.1162/neco_a_01096
- Chicago author-date
- Stock, Michiel, Tapio Pahikkala, Antti Airola, Bernard De Baets, and Willem Waegeman. 2018. “A Comparative Study of Pairwise Learning Methods Based on Kernel Ridge Regression.” NEURAL COMPUTATION 30 (8): 2245–83. https://doi.org/10.1162/neco_a_01096.
- Chicago author-date (all authors)
- Stock, Michiel, Tapio Pahikkala, Antti Airola, Bernard De Baets, and Willem Waegeman. 2018. “A Comparative Study of Pairwise Learning Methods Based on Kernel Ridge Regression.” NEURAL COMPUTATION 30 (8): 2245–2283. doi:10.1162/neco_a_01096.
- Vancouver
- 1.Stock M, Pahikkala T, Airola A, De Baets B, Waegeman W. A comparative study of pairwise learning methods based on kernel ridge regression. NEURAL COMPUTATION. 2018;30(8):2245–83.
- IEEE
- [1]M. Stock, T. Pahikkala, A. Airola, B. De Baets, and W. Waegeman, “A comparative study of pairwise learning methods based on kernel ridge regression,” NEURAL COMPUTATION, vol. 30, no. 8, pp. 2245–2283, 2018.
@article{8578969,
abstract = {{Many machine learning problems can be formulated as predicting labels for a pair of objects. Problems of that kind are often referred to as pairwise learning, dyadic prediction, or network inference problems. During the past decade, kernel methods have played a dominant role in pairwise learning. They still obtain a state-of-the-art predictive performance, but a theoretical analysis of their behavior has been underexplored in the machine learning literature. In this work we review and unify kernel-based algorithms that are commonly used in different pairwise learning settings, ranging from matrix filtering to zero-shot learning. To this end, we focus on closed-form efficient instantiations of Kronecker kernel ridge regression. We show that independent task kernel ridge regression, two-step kernel ridge regression, and a linear matrix filter arise naturally as a special case of Kronecker kernel ridge regression, implying that all these methods implicitly minimize a squared loss. In addition, we analyze urdversality, consistency, and spectral filtering properties. Our theoretical results provide valuable insights into assessing the advantages and limitations of existing pairwise learning methods.}},
author = {{Stock, Michiel and Pahikkala, Tapio and Airola, Antti and De Baets, Bernard and Waegeman, Willem}},
issn = {{0899-7667}},
journal = {{NEURAL COMPUTATION}},
keywords = {{DRUG-TARGET INTERACTIONS,SUPPORT VECTOR MACHINES,LEAST-SQUARES,REGULARIZATION ALGORITHMS,INTERACTION PREDICTION,MATRIX FACTORIZATION,NETWORKS}},
language = {{eng}},
number = {{8}},
pages = {{2245--2283}},
title = {{A comparative study of pairwise learning methods based on kernel ridge regression}},
url = {{http://doi.org/10.1162/neco_a_01096}},
volume = {{30}},
year = {{2018}},
}
- Altmetric
- View in Altmetric
- Web of Science
- Times cited: