Bayesian Quadrature for Neural Ensemble Search
This work addresses a specific problem in neural network ensembling for machine learning practitioners, offering an incremental improvement over existing methods.
The paper tackled the problem of constructing neural network ensembles when architecture likelihood surfaces have dispersed, narrow peaks, and existing methods use equally weighted ensembles vulnerable to weaker architectures. The result is a method using Bayesian Quadrature to weight architectures by performance, empirically outperforming state-of-the-art baselines in test likelihood, accuracy, and expected calibration error.
Ensembling can improve the performance of Neural Networks, but existing approaches struggle when the architecture likelihood surface has dispersed, narrow peaks. Furthermore, existing methods construct equally weighted ensembles, and this is likely to be vulnerable to the failure modes of the weaker architectures. By viewing ensembling as approximately marginalising over architectures we construct ensembles using the tools of Bayesian Quadrature -- tools which are well suited to the exploration of likelihood surfaces with dispersed, narrow peaks. Additionally, the resulting ensembles consist of architectures weighted commensurate with their performance. We show empirically -- in terms of test likelihood, accuracy, and expected calibration error -- that our method outperforms state-of-the-art baselines, and verify via ablation studies that its components do so independently.