Training Minimum Enclosing Balls for Cross Tasks Knowledge Transfer

Abstract

This paper proposes a learner independent multi-task learning (MTL) scheme such that $\mathcal{M}_\mathcal{L} = \mathcal{L}(T^i, KT(T^i, T^j))$, for i,j = 1,2, i ≠ j, where KT is independent to the learner $\mathcal{L}$, and MTL is conducted for arbitrary learner combinations. In the proposed solution, we use Minimum Enclosing Balls (MEBs) as knowledge carriers to extract and transfer knowledge from one task to another. Since the knowledge presented in MEB can be decomposed as raw data, it can be incorporated into any learner as additional training data for a new learning task and thus improve its learning rate. The effectiveness and robustness of the proposed KT is evaluated on multi-task pattern recognition (MTPR) problems derived from UCI datasets, using classifiers from different disciplines for MTL. The experimental results show that multi-task learners using KT via MEB carriers perform better than learners without-KT, and it is successfully applied to all type of classifiers.