Loading
Intelligent agents, search algorithms, constraint satisfaction, adversarial search, logic and planning, probabilistic reasoning, and decision making under uncertainty — the classical foundations of AI.
Other tracks: Quantitative Finance · Software Engineering
The agent–environment loop: percepts, actions, and what makes an agent rational.
What it means for an agent to be rational, and why specifying the right objective is the hard part.
States, actions, transitions, costs, goals — the formal vocabulary of problem solving in AI.
Time, space, completeness, optimality — the four properties every search algorithm is judged on.
Explore layer by layer — complete, optimal for uniform cost, expensive in memory.
Go deep before going wide — cheap memory, no guarantees.
BFS generalized to non-uniform costs — the right uninformed algorithm for weighted graphs.
Run DFS with growing depth limits — get BFS's completeness at DFS's memory cost.
Domain knowledge that estimates remaining cost — admissibility, consistency, and dominance.
Best-first search guided by $f(n) = g(n) + h(n)$ — the most important informed search algorithm.
Iterative deepening with cost cutoffs — A*'s pruning power at DFS's memory cost.
Keep only the $k$ best frontier nodes — fast and bounded, but no longer optimal.
Variables, domains, constraints — and why CSP solvers solve such a wide range of problems.
Depth-first assignment with constraint checking — the foundation under every CSP solver.
Propagate constraints across the variable graph until no value can be eliminated.
Min-conflicts and simulated annealing — escape backtracking when the state space is too large to enumerate.
Two-player zero-sum games — the algorithm at the heart of every classical game AI.
Skip subtrees that can't change the answer — the optimization that made minimax viable for real games.
Games with chance — backgammon, dice games, poker (partially).
Repeatedly simulate to leaf, update statistics — the algorithm that broke Go.
Atomic propositions, connectives, entailment — the foundation of symbolic reasoning.
Add objects, predicates, and quantifiers — express "every" and "there exists".
The single proof rule that proves anything in first-order logic — when combined with skolemization.
STRIPS, PDDL, and the algorithms that turn goals into action sequences.
Compact representations of joint distributions via conditional independence.
The standard exact inference algorithm — and why ordering choices matter.
Sequential models with hidden state — the workhorse before deep learning.
When exact inference is impossible — Monte Carlo methods that scale to messy real problems.
States, actions, rewards, transitions — the framework underneath reinforcement learning.
Iterate the Bellman update until convergence — the simplest exact MDP solver.
Model-free reinforcement learning — learn $Q^*$ from samples without ever knowing the transition model.
Learn the policy directly — REINFORCE, PPO, and the algorithms behind modern RL.