Importance Weighted Inductive Transfer Learning for Regression
Abstract
We consider inductive transfer learning for dataset shift, a situation in which the distributions of two sampled, but closely related, datasets differ. When the target data to be predicted is scarce, one would like to improve its prediction by employing data from the other, secondary, dataset. Transfer learning tries to address this task by suitably compensating such a dataset shift. In this work we assume that the distributions of the covariates and the dependent variables can differ arbitrarily between the datasets. We propose two methods for regression based on importance weighting. Here to each instance of the secondary data a weight is assigned such that the data contributes positively to the prediction of the target data. Experiments show that our method yields good results on benchmark and real world datasets.
Cite
Text
Garcke and Vanck. "Importance Weighted Inductive Transfer Learning for Regression." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2014. doi:10.1007/978-3-662-44848-9_30Markdown
[Garcke and Vanck. "Importance Weighted Inductive Transfer Learning for Regression." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2014.](https://mlanthology.org/ecmlpkdd/2014/garcke2014ecmlpkdd-importance/) doi:10.1007/978-3-662-44848-9_30BibTeX
@inproceedings{garcke2014ecmlpkdd-importance,
title = {{Importance Weighted Inductive Transfer Learning for Regression}},
author = {Garcke, Jochen and Vanck, Thomas},
booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases},
year = {2014},
pages = {466-481},
doi = {10.1007/978-3-662-44848-9_30},
url = {https://mlanthology.org/ecmlpkdd/2014/garcke2014ecmlpkdd-importance/}
}