Improving robustness and calibration in ensembles with diversity regularization

TL;DR

This paper proposes a new diversity regularizer for ensemble classifiers that improves calibration, robustness, and out-of-distribution detection by leveraging out-of-distribution samples and explicitly promoting diversity among ensemble members.

Contribution

It introduces a novel diversity regularizer that enhances ensemble performance and calibration, especially in architectures with shared weights, and demonstrates its effectiveness on multiple datasets.

Findings

Regularizing diversity improves calibration and robustness.

Diversity regularization enhances out-of-distribution detection.

Fewer ensemble members are needed to achieve robustness.

Abstract

Calibration and uncertainty estimation are crucial topics in high-risk environments. We introduce a new diversity regularizer for classification tasks that uses out-of-distribution samples and increases the overall accuracy, calibration and out-of-distribution detection capabilities of ensembles. Following the recent interest in the diversity of ensembles, we systematically evaluate the viability of explicitly regularizing ensemble diversity to improve calibration on in-distribution data as well as under dataset shift. We demonstrate that diversity regularization is highly beneficial in architectures, where weights are partially shared between the individual members and even allows to use fewer ensemble members to reach the same level of robustness. Experiments on CIFAR-10, CIFAR-100, and SVHN show that regularizing diversity can have a significant impact on calibration and robustness, as well as out-of-distribution detection.