ICML 2011 workshop on unsupervised and transfer learning

Transfer Learning by Kernel Meta-Learning
Fabio Aiolli
University of Padova, Italy

A crucial issue in machine learning is how to find good representations for data. Recently, much work has been devoted to kernel learning, that is the problem to find a good kernel matrix for a task. This can be made in a semi-supervised learning setting by using a large set of unlabeled data and a (typically small) set of i:i:d: labeled data. Another, even more challenging problem, is how one could exploit partially labeled data for a source task to learn good representations for another related but dierent target task. This is the main subject of transfer learning. In this paper, we present a novel approach to transfer learning based on kernel learning. Specifically, we propose a kernel meta-learning algorithm which, starting from a basic kernel, tries to learn chains of kernel transforms that are able to produce good kernel matrices for the source tasks. The same sequence of transformations can be then applied to learn the kernel matrix for any target task. We report on the application of this method to the five datasets of the Unsupervised and Transfer Learning (UTL) challenge benchmark. Nicely, this technique allowed us to win the first phase of the competition.