Pool-based sequential active learning with multi kernels
This work addresses the problem of efficient sample selection in active learning for researchers and practitioners, but it appears incremental as it builds on and generalizes established criteria.
The paper tackled pool-based sequential active learning by proposing two new selection criteria, EKD and EKL, based on multiple kernel learning, which generalize existing methods like QBC and EMC. Experimental results on real datasets verified their effectiveness compared to existing methods, though no specific numerical gains were reported.
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.