A Hill-Climbing Landmarker Generation Algorithm Based on Efficiency and Correlativity Criteria

Daren Ler, Irena Koprinska, and Sanjay Chawla, University of Sydney

For a given classification task, there are typically several learning algorithms available. The question then arises: which is the most appropriate algorithm to apply. Recently, we proposed a new algorithm for making such a selection based on landmarking - a meta-learning strategy that utilises meta-features that are measurements based on efficient learning algorithms. This algorithm, which creates a set of landmarkers that each utilise subsets of the algorithms being landmarked, was shown to be able to estimate accuracy well, even when employing a small fraction of the given algorithms. However, that version of the algorithm has exponential computational complexity for training. In this paper, we propose a hill-climbing version of the landmarker generation algorithm, which requires only polynomial training time complexity. Our experiments show that the landmarkers formed have similar results to the more complex version of the algorithm.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.