Improving greedy core-set configurations for active learning with uncertainty-scaled distances

TL;DR

This paper enhances core-set based active learning by incorporating uncertainty-scaled distances, leading to significant improvements in sample efficiency for image classification tasks.

Contribution

It introduces a novel uncertainty scaling method for core-set configurations, improving sample efficiency and convergence speed in active learning.

Findings

Significant sample efficiency improvements on CIFAR10/100 and SVHN.

Probabilistic quadratic speed-up in core-set loss convergence.

Necessity of the proposed modifications demonstrated.

Abstract

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.