Web-Induced Heterogeneous Transfer Learning with Sample Selection
Abstract
Transfer learning algorithms utilize knowledge from a data-rich source domain to learn a model in the target domain where labeled data is scarce. This paper presents a novel solution for the challenging and interesting problem of Heterogeneous Transfer Learning (HTL) where the source and target task have heterogeneous feature and label spaces. Contrary to common space based HTL algorithms, the proposed HTL algorithm adapts source data for the target task. The correspondence required for aligning the heterogeneous features of the source and target domain is obtained through labels across two domains that are semantically aligned using web-induced knowledge. The experimental results suggest that the proposed algorithm performs significantly better than state-of-the-art transfer approaches on three diverse real-world transfer tasks.
Cite
Text
Sukhija and Krishnan. "Web-Induced Heterogeneous Transfer Learning with Sample Selection." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2018. doi:10.1007/978-3-030-10928-8_46Markdown
[Sukhija and Krishnan. "Web-Induced Heterogeneous Transfer Learning with Sample Selection." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2018.](https://mlanthology.org/ecmlpkdd/2018/sukhija2018ecmlpkdd-webinduced/) doi:10.1007/978-3-030-10928-8_46BibTeX
@inproceedings{sukhija2018ecmlpkdd-webinduced,
title = {{Web-Induced Heterogeneous Transfer Learning with Sample Selection}},
author = {Sukhija, Sanatan and Krishnan, Narayanan Chatapuram},
booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases},
year = {2018},
pages = {777-793},
doi = {10.1007/978-3-030-10928-8_46},
url = {https://mlanthology.org/ecmlpkdd/2018/sukhija2018ecmlpkdd-webinduced/}
}