NES Surrogate Architectures: Trained Neural Networks for Ensemble Search

A collection of trained neural architectures for surrogate-assisted neural ensemble search (NES). The dataset includes structural definitions, model weights, validation predictions, and validation accuracy for models trained on CIFAR10, CIFAR100, and FashionMNIST. It was authored by Demoren and last updated on May 1, 2026.

Use Cases

• Study performance prediction of neural architectures based on their structural definitions and validation accuracy.
• Analyze model diversity estimation for ensemble methods using the provided validation predictions.
• Benchmark surrogate-assisted search algorithms using the pre-trained architectures and associated weights.
• Investigate the relationship between DARTS-like cell structures and model performance on standard computer vision datasets.

Strengths

• Includes multiple components per architecture: structural definition, model weights, validation predictions, and validation accuracy.
• Covers three standard computer vision benchmark datasets: CIFAR10, CIFAR100, and FashionMNIST.
• Last updated on 2026-05-01 15:36:27, indicating recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Collection of trained neural architectures, likely generated via neural architecture search processes.

Time Range

null

Freshness

Last updated 2026-05-01 15:36:27; freshness should be verified.

Geography

null