Pool-based sequential active learning with multi kernels

TL;DR

This paper introduces two novel selection criteria, EKD and EKL, for pool-based sequential active learning that leverage multiple kernel learning structures, demonstrating improved effectiveness over existing methods.

Contribution

The paper proposes EKD and EKL criteria for active learning, generalizing QBC and EMC, tailored for multiple kernel learning frameworks.

Findings

EKD and EKL outperform existing active learning methods

Experimental results validate the effectiveness of the proposed criteria

EKD and EKL successfully generalize popular query strategies

Abstract

We study a pool-based sequential active learning (AL), in which one sample is queried at each time from a large pool of unlabeled data according to a selection criterion. For this framework, we propose two selection criteria, named expected-kernel-discrepancy (EKD) and expected-kernel-loss (EKL), by leveraging the particular structure of multiple kernel learning (MKL). Also, it is identified that the proposed EKD and EKL successfully generalize the concepts of popular query-by-committee (QBC) and expected-model-change (EMC), respectively. Via experimental results with real-data sets, we verify the effectiveness of the proposed criteria compared with the existing methods.