Every concept is a node; every arrow is a relationship. Follow the dark spine to see how the course evolves, watch each lecture branch into its ideas, and trace the dashed threads to see how one idea gets reused later. Tap a node for its definition · tap a thread to follow it across the course.
Click a concept — say Expected utility or Nash equilibrium — and this box gives you an exam-ready definition, the formula where there is one, a plain-words line, and a link to its full lecture. The dark arrows along the top are the story of the course; the dashed lines are ideas that come back later.
The lectures aren't separate islands — a few ideas reappear, getting richer each time. Tap a card to light its thread up on the map.
The value of an outcome — the thing every decision tries to maximise.
Real decisions are never deterministic — we reason over probability distributions.
You're rarely deciding alone; other goal-seeking agents interact with yours.
The fact that flips the course from clean theory to real systems.
A mathematically optimal answer is useless if the human won't accept or act on it.
The whole arc, in one breath — the order to keep straight for the exam.
The single sentence to be able to say out loud for each lecture.
| # | Lecture | The one thing |
|---|---|---|
| 01 | Decision Theory | A rational choice maximises expected utility; under pure ignorance you instead pick a rule (maximin, maximax, minimax-regret). |
| 02 | Bayesian & Decision Networks | Networks make a huge joint distribution compact via conditional independence — but exact inference is still #P-hard. |
| 03 | Game Theory | A Nash equilibrium is a mutual best response; it always exists (maybe mixed), but isn't always efficient (prisoner's dilemma). |
| 04 | Sequential Decisions | Know the model → dynamic programming; no model → reinforcement learning; have a simulator → MCTS. Bellman ties them together. |
| 05 | Behavioral | Humans use heuristics and have systematic biases — yet heuristics can be “ecologically rational”; prospect theory values gains/losses around a reference point. |
| 06 | Architecture | A DSS augments a human (≠ replaces); separate data / knowledge / model / interface; human factors are design constraints. |
| 07 | Evaluation | Good DSS ≠ good predictor: judge accuracy + calibration + fairness + robustness + usability + utility + the human-DSS team. |
Two halves to keep straight: the professor said the course moves from normative decision theory (what we should do, Lectures 1–4) through the behavioral reality (Lecture 5) to the design and evaluation of the support systems themselves (Lectures 6–7).