Architecture as an emergent property of neural network training: Dataset and Experiments

This archive contains all datasets used in the ASANN (Autonomously Self-Architecting Neural Network) study, organised by experimental tier. The data support more than seventy-five experiments spanning seven computational domains: tabular regression and classification, image classification, graph and spatio-temporal learning, molecular property prediction, physics-informed neural networks for partial differential equations, pharmacogenomic drug-response modelling, and haematological cell classification. All experiments use a fixed 70/15/15 train/validation/test split with stratification for classification tasks and scaffold splitting for molecular benchmarks. No dataset was modified, augmented, or filtered beyond the standardisation described in the Methods section of the accompanying article.