Improving greedy core-set configurations for active learning with uncertainty-scaled distances
This work addresses sample efficiency for active learning practitioners, though it appears incremental as it modifies an existing core-set algorithm.
The paper tackles the problem of improving sample efficiency in active learning by scaling core-set distances with model uncertainty and searching for low-confidence configurations, achieving significant improvements on CIFAR10/100 and SVHN datasets, particularly with larger acquisition sizes.
We scale perceived distances of the core-set algorithm by a factor of uncertainty and search for low-confidence configurations, finding significant improvements in sample efficiency across CIFAR10/100 and SVHN image classification, especially in larger acquisition sizes. We show the necessity of our modifications and explain how the improvement is due to a probabilistic quadratic speed-up in the convergence of core-set loss, under assumptions about the relationship of model uncertainty and misclassification.