Section 2

Learning Paradigms

These five Beginner chapters open the learning side of AI: a churn model trained by gradient descent and broken by leakage, supervised evaluation on real tumour data and a simulated fraud screen, clustering and embeddings, reinforcement learning from bandits to the post-training of language models, and deep networks with their memory and compute bills. Every number is printed by a Python and Java program or cited from a published result, from Google Flu Trends and a pneumonia model that recognised hospitals to AlexNet's 15.3% ImageNet error.