Classical Artificial Intelligence

Rational agents, PEAS descriptions, and the environment taxonomy: observable, deterministic, episodic, discrete.

States, actions, transition models, goal tests and path costs, with the state-space graph.

Assumes: Agents and Environments

BFS, DFS, uniform-cost, depth-limited and iterative deepening, compared on completeness, optimality and complexity.

Assumes: Problem Formulation and State Spaces · Breadth-First and Depth-First Search

Greedy best-first and A*, the f = g + h decomposition, and a full numeric trace of A* on a map.

Assumes: Uninformed Search

Designing heuristics, admissibility and consistency proofs, dominance, and relaxed-problem heuristics.

Assumes: Informed Search and A*

Hill climbing and its traps, simulated annealing, beam search, and genetic algorithms.

Assumes: Informed Search and A*

Turning the minimax theorem into an algorithm: game-tree search, the minimax value, and a fully traced example.

Assumes: Problem Formulation and State Spaces · Zero-Sum Games and the Minimax Theorem

Pruning without changing the result, move-ordering effects, and expectimax for stochastic games.

Assumes: Adversarial Search and Minimax

Variables, domains and constraints; backtracking, forward checking, arc consistency and AC-3.

Assumes: Local Search and Metaheuristics

Semantic networks, frames, scripts and ontologies, and the trade-off between expressiveness and tractable inference.

Assumes: First-Order Logic: Syntax

Knowledge bases, TELL and ASK, and an agent deducing safe moves from percepts.

Assumes: Resolution and Refutation · Agents and Environments

STRIPS representation, state-space and plan-space search, partial-order planning and GraphPlan.

Assumes: Knowledge Representation · Informed Search and A*

Why logical entailment is too brittle for real evidence, the probabilistic alternative, and full joint distributions with their cost.

Assumes: Bayes' Theorem · Validity, Satisfiability and Entailment

DAG structure, conditional probability tables, factorisation of the joint, and constructing a network.

Assumes: Reasoning Under Uncertainty

Chains, forks and colliders; reading independence off the graph and the explaining-away effect.

Assumes: Bayesian Networks

Factor products and summing out, elimination ordering, and a complete worked inference.

Assumes: Conditional Independence and d-Separation

Prior, rejection and likelihood-weighted sampling, then Gibbs sampling and MCMC convergence.

Assumes: Exact Inference by Variable Elimination

States, observations, the three canonical problems, and HMMs as dynamic Bayesian networks.

Assumes: Markov Chains · Bayesian Networks

Filtering, smoothing and the most likely state sequence, each traced numerically on a small HMM.

Assumes: Hidden Markov Models