Heterogeneous Few-Shot Model Rectification With Semantic Mapping
Lists
Only you see this
Abstract
There still involve lots of challenges when applying machine learning algorithms in <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">unknown</i> environments, especially those with <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">limited</i> training data. To handle the data insufficiency and make a further step towards robust learning, we adopt the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">learnware</i> notion Z.-H. Zhou, “Learnware: On the future of machine learning,” <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Front. Comput. Sci.</i> , vol. 10, no. 4 pp. 589–590, 2016 which equips a model with an essential <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">reusable</i> property—the model learned in a related task could be easily adapted to the current data-scarce environment <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">without data sharing</i> . To this end, we propose the REctiFy via heterOgeneous pRedictor Mapping ( <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ReForm</small> ) framework enabling the current model to take advantage of a related model from two kinds of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">heterogeneous environment</i> , i.e., either with different sets of features or labels. By Encoding Meta InformaTion ( <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Emit</small> ) of features and labels as the model specification, we utilize an optimal transported semantic mapping to characterize and bridge the environment changes. After fine-tuning over a few labeled examples through a biased regularization objective, the transformed heterogeneous model adapts to the current task efficiently. We apply <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ReForm</small> over both synthetic and real-world tasks such as few-shot image classification with either learned or pre-defined specifications. Experimental results validate the effectiveness and practical utility of the proposed <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ReForm</small> framework.
BibTeX
@article{ye2021heterogeneous,
author = {Han-Jia Ye and De-Chuan Zhan and Yuan Jiang and Zhi-Hua Zhou},
title = {{Heterogeneous Few-Shot Model Rectification With Semantic Mapping}},
journal = {IEEE Trans. Pattern Anal. Mach. Intell.},
year = {2021},
url = {https://openreview.net/forum?id=5FnCcIkwEd},
}Loading related content...