Represent knowledge and learn from data

Choose a reasoning model

A rule-based system applies explicit rules to stored facts. A search algorithm explores candidate states according to a strategy. A heuristic estimates which choices are promising, but its usefulness depends on the problem and on any guarantees the algorithm needs.

Representing a problem well matters as much as selecting a procedure. State what a node means, what transitions are allowed, and how a goal is recognized. A diagram without those meanings is not a complete search model.

Evaluate a learned model

Supervised learning uses labeled examples, unsupervised learning looks for structure without supplied targets, and reinforcement learning uses feedback from actions. A neural network adjusts weights during training; it is not a collection of manually written answers for every possible input.

Keep evaluation data separate from training, inspect incorrect predictions, and consider bias and consequences. High training accuracy can indicate memorization instead of reliable behavior on unseen data. Explain whether false positives or false negatives are more costly in the scenario.

Check yourself