Neural Network Foundations | From Computation to Responsible Evidence | SRAI Book 3 Lesson 1

This university-level SRAI lesson builds neural networks from the artificial neuron upward, then connects architecture to reproducible and accountable evidence. A controlled synthetic public-service case demonstrates forward propagation, activation functions, data partitioning, baseline comparison, gradient checking, checkpoint selection, calibration, thresholds, subgroup analysis, interpretation, distribution shift, reproducibility and model documentation. The results are educational and synthetic. They do not establish causal effects, real-population validity, fairness, deployment safety or authority for automated decisions. Production unit: PU-B03-C01 Canonical notebook: M3_N01_neural_network_foundations SRAI — From Statistical Reasoning to Responsible Artificial Intelligence