Multi-headed Neural Ensemble Search

TL;DR

This paper introduces a multi-headed neural ensemble search method that enables efficient end-to-end training of diverse ensembles, significantly reducing search time while maintaining high performance and uncertainty calibration.

Contribution

It extends neural ensemble search to multi-headed architectures, allowing one-shot NAS optimization and faster ensemble discovery without sacrificing accuracy.

Findings

Multi-headed ensembles are trained end-to-end, enabling efficient search.

The proposed method finds robust ensembles 3 times faster than traditional methods.

Performance and uncertainty calibration are comparable to existing ensemble search techniques.

Abstract

Ensembles of CNN models trained with different seeds (also known as Deep Ensembles) are known to achieve superior performance over a single copy of the CNN. Neural Ensemble Search (NES) can further boost performance by adding architectural diversity. However, the scope of NES remains prohibitive under limited computational resources. In this work, we extend NES to multi-headed ensembles, which consist of a shared backbone attached to multiple prediction heads. Unlike Deep Ensembles, these multi-headed ensembles can be trained end to end, which enables us to leverage one-shot NAS methods to optimize an ensemble objective. With extensive empirical evaluations, we demonstrate that multi-headed ensemble search finds robust ensembles 3 times faster, while having comparable performance to other ensemble search methods, in both predictive performance and uncertainty calibration.