| Ensembles are a standard way to improve the performance and robustness of deep neural networks, but their effectiveness crucially depends on both the quality and diversity of individual models. Most neural architecture search (NAS) methods are computationally expensive, and extending them to neural ensemble search (NES), where both individual architectures and ensemble composition must be optimized jointly, leads to exponential growth of the search space and makes the problem computationally intractable. To address this challenge, we introduce a dual-objective surrogate-guided ensemble search framework. Candidate architectures are represented as directed acyclic graphs, and two surrogate models are trained independently to estimate predictive accuracy and diversity potential. Their combined estimates guide the search toward architectures that are both individually strong and collectively diverse. The resulting ensembles achieve competitive or superior performance compared to standard baselines, including Deep Ensembles and Random Search, on FashionMNIST, CIFAR-10, and CIFAR-100. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.