Learning to learn (L2L) trains a meta-learner to assist the learning of a task-specific base learner. Previously, it was shown that a meta-learner could learn the direct rules to update learner parameters; and that the learnt neural optimiser updated learners more rapidly than handcrafted gradient-descent methods. However, we demonstrate that previous neural optimisers were limited to update learners on one designated dataset. In order to address input-domain heterogeneity, we introduce Multi-Task Learning to Learn (MTL2L), a context aware neural optimiser which self-modifies its optimisation rules based on input data. We show that MTL2L is capable of updating learners to classify on data of an unseen input-domain at the meta-testing phase.
@article{arxiv.2007.09343,
title = {MTL2L: A Context Aware Neural Optimiser},
author = {Nicholas I-Hsien Kuo and Mehrtash Harandi and Nicolas Fourrier and Christian Walder and Gabriela Ferraro and Hanna Suominen},
journal= {arXiv preprint arXiv:2007.09343},
year = {2020}
}
Comments
Published in the ICML workshop of Automated Machine Learning (AutoML) 2020. Also see https://www.automl.org/wp-content/uploads/2020/07/AutoML_2020_paper_5.pdf