DICE: Diversity in Deep Ensembles via Conditional Redundancy Adversarial Estimation

TL;DR

DICE introduces a novel training method for deep ensembles that balances diversity and individual accuracy by reducing feature redundancy through adversarial training, leading to improved ensemble performance.

Contribution

It proposes a new criterion called DICE that leverages conditional mutual information estimation and adversarial training to enhance diversity without sacrificing individual network accuracy.

Findings

Achieves state-of-the-art accuracy on CIFAR-10/100 datasets.

Ensemble of 5 networks with DICE matches the performance of 7 independently trained networks.

Improves calibration, uncertainty estimation, and out-of-distribution detection.

Abstract

Deep ensembles perform better than a single network thanks to the diversity among their members. Recent approaches regularize predictions to increase diversity; however, they also drastically decrease individual members' performances. In this paper, we argue that learning strategies for deep ensembles need to tackle the trade-off between ensemble diversity and individual accuracies. Motivated by arguments from information theory and leveraging recent advances in neural estimation of conditional mutual information, we introduce a novel training criterion called DICE: it increases diversity by reducing spurious correlations among features. The main idea is that features extracted from pairs of members should only share information useful for target class prediction without being conditionally redundant. Therefore, besides the classification loss with information bottleneck, we adversarially prevent features from being conditionally predictable from each other. We manage to reduce simultaneous errors while protecting class information. We obtain state-of-the-art accuracy results on CIFAR-10/100: for example, an ensemble of 5 networks trained with DICE matches an ensemble of 7 networks trained independently. We further analyze the consequences on calibration, uncertainty estimation, out-of-distribution detection and online co-distillation.