COMP9414 Artificial Intelligence
The Complete Exam Bible · T3 2026COMP9414 Overview
- UNSW Sydney
- Term 3 2026
- Postgraduate course
- Double-pass assessment
COMP9414 Artificial Intelligence is a postgraduate UNSW Sydney course offered in Term 3, 2026. The official schedule moves from agents and search through planning, knowledge representation, neural networks, reinforcement learning, reasoning with uncertainty, computer vision, foundation models and human-aligned intelligent robotics.
- Assessment Two assignments, tutorial work and a final exam determine the course mark.
- Hurdle Reach the published exam threshold as well as the overall pass mark.
- Hardest move Keep state, path history, priority and guarantee in separate columns.
- Planning check Stop at the first false precondition before attempting a repair.
What you get in the full COMP9414 Exam Bible
- 5 chapters, 30 pages — every examinable topic, week by week
- 16 original practice questions, each with a model answer in the Final exam format
- 5 worked examples, solved the way COMP9414 marks
- 5 computed figures, redrawn from the course’s own numbers
- Exam-morning plan for the Final exam (50%) — what to revise last, where marks are lost
Free on this page: assessment table, chapter map, one worked example, key terms, FAQ. The rest is in the full exam bible.
How COMP9414 is assessed
| Component | Weight | Format |
|---|---|---|
| Assignment 1 | 20% | Individual; issued Week 3 and due Week 5 |
| Assignment 2 | 20% | Individual; issued Week 7 and due Week 9 |
| Tutorial work | 10% | Individual work assessed each week |
| Final exam · hurdle | 50% | Individual; examination period; exam mark must reach 20 out of 50 |
What other students are asking about COMP9414
- Where can I find lecture notes or study guides for COMP9414 at UNSW?
- The COMP9414 study guide in AskSia Library covers Artificial Intelligence in 5 chapters, with the assessment table, a chapter-by-chapter topic map, key terms and a worked example, free to read on this page.
- Is there a final exam in COMP9414, and how much is it worth?
- Yes. In the offering this AskSia Library guide is built from, Final exam carries 50% of the COMP9414 mark; the full split is Assignment 1 20%, Assignment 2 20%, Tutorial work 10%, Final exam 50%.
- What does COMP9414 cover on rational agents and task models?
- In AskSia Library's COMP9414 guide, the Rational Agents and Task Models chapter explains define percepts, actions, environment properties, goals, costs and utility.
- What does COMP9414 cover on states, trees and frontiers?
- In AskSia Library's COMP9414 guide, the States, Trees and Frontiers chapter explains formulate search problems and keep state graphs distinct from path histories.
- What does COMP9414 cover on search strategy and heuristic choice?
- In AskSia Library's COMP9414 guide, the Search Strategy and Heuristic Choice chapter explains compare stack, queue, path-cost and estimated-total-cost frontier rules.
What COMP9414 covers
The detailed teaching materials available for this offering build a connected foundation: evaluate claims about intelligent behaviour, model an agent and its environment, formulate a state-space problem, control the frontier with the right priority, and make planning actions explicit through preconditions and effects.
AI as Evidence and Capability
Separate observed behaviour, mechanism, task boundary and a falsifiable capability claimThe course history uses apparent autonomy, conversational behaviour, bounded game success, embodied agents and combinations of learning with search to ask a harder question than whether a demonstration looks impressive.
- Observed behaviour versus an interpretation about intelligence
- Mechanism evidence from search, learning or integrated agent design
- Bounded success in a precisely defined task
- Embodied agents facing perceptual noise and other agents
- Combined methods such as learning with tree search
- Boundary tests that change context, assumptions or task distribution
Rational Agents and Task Models
Define percepts, actions, environment properties, goals, costs and utilityAn agent perceives, decides and acts in an environment, but those verbs become computational only after the model names percepts, legal actions, relevant world features and the preferred outcome. This chapter turns that loop into a task specification.
- Percepts entering an agent from its environment
- Actions changing the world and future observations
- Fully and partially observable task models
- Deterministic and stochastic action outcomes
- Static and dynamic environments
- Goals, path costs and utility as different roles
States, Trees and Frontiers
Formulate search problems and keep state graphs distinct from path historiesSearch depends on a formulation before it depends on a data structure. This chapter defines the initial state, successor function, goal test and path cost, then separates the state-space graph from the search tree generated from a start.
- Initial state, successors, goal test and path cost
- State-space graphs with one node per world state
- Search trees that unfold alternative plan histories
- Frontier entries with parents, depth and cumulative cost
- Discovery, expansion and returned-path events
- Repeated-state handling under weighted and unweighted search
Search Strategy and Heuristic Choice
Compare stack, queue, path-cost and estimated-total-cost frontier rulesDFS, BFS, UCS and A* share a search loop but assign different meanings to the best frontier candidate. This chapter derives their guarantees from stack, queue, cumulative-cost and estimated-total-cost priorities.
- Depth-first search with a LIFO frontier
- Breadth-first search with a FIFO frontier
- Uniform-cost search ordered by cumulative g
- Greedy search ordered by heuristic h
- A* search ordered by f equal to g plus h
- Admissibility as an optimistic remaining-cost estimate
STRIPS Planning and Goal Interaction
Ground actions, test preconditions, update facts and repair interacting goalsPlanning opens the successor function by representing states as fact sets and actions through preconditions, add effects and delete effects.
- Closed-world symbolic states and ground facts
- Operator schemas and grounded actions
- Preconditions as an applicability gate
- Add and delete effects in the state transition
- Forward state-space planning with BFS
- Interacting goals, monitoring and replanning
The detailed teaching files available for this guide concentrate on the first connected foundation: how to judge evidence about intelligent behaviour, how to describe an agent and its environment, how to formulate a state-space problem, how frontier discipline changes search behaviour, and how STRIPS makes action preconditions and effects explicit. The assessment structure places substantial pressure on reliable execution.
Assignment 1 contributes 20%, Assignment 2 contributes 20%, tutorial work contributes 10%, and the Final exam contributes 50%. The course also states a double-pass rule: the overall mark must reach 50 out of 100 and the exam mark must reach 20 out of 50. The most useful revision habit is therefore to show a trace that can be checked.
For search, keep the frontier, path, depth and cumulative cost distinct, then write g, h and f separately for A*. For planning, ground each action, test all preconditions, apply delete and add effects, preserve unrelated facts and verify the goal after the last step. This guide uses original graphs and scenarios modelled on those question types.
It does not reproduce the course exercises or imply a past exam format that the captured materials do not state.
Worked example · free
A* trace with an improved route
- +1From S, A has g=2 and f=6; B has g=4 and f=6. Apply the stated tie rule and expand A.
- +1From A, C has g=4 and f=5 while the direct goal has g=8 and f=8, so C becomes the next expansion.
- +1From C, improve the goal to g=6 and f=6. A later route from B to C would cost 5, so it does not improve C.
- +1Return S to A to C to G with actual cost 2+2+2=6. The heuristic orders the frontier but is not added to the route cost.
Key terms
- Agent
- A decision-making system that receives percepts and can affect its surroundings through actions.
- Environment
- The external world, objects and other agents that can affect a task or outcome.
- State space
- The set of world states and legal transitions defined by a problem formulation.
- Search tree
- The unfolding of alternative paths from a start state, where a world state may appear under several histories.
- Frontier
- The collection of generated partial paths waiting to be selected for expansion.
- Completeness
- A guarantee that a method finds a solution when one exists under its stated conditions.
- Optimality
- A guarantee that the returned solution minimises the defined path cost under stated assumptions.
- Heuristic
- An estimate of remaining cost used to guide search toward a goal.
- Admissible heuristic
- A heuristic that never overestimates the true cheapest remaining cost.
- Grounded action
- An action schema whose variables have all been replaced by concrete objects.
- Delete effect
- A fact made false when an applicable planning action is executed.
- Goal interaction
- A dependency where achieving one goal changes a condition needed to achieve or preserve another.
COMP9414 FAQ
How is the course assessed?
The published Term 3 table assigns 20% to Assignment 1, 20% to Assignment 2, 10% to tutorial work and 50% to the Final exam. The numeric rows total 100%. Each assignment is individual, and the table places them in the middle and later parts of the teaching term.
What is the exam hurdle?
Passing requires both an overall mark of at least 50 out of 100 and an exam mark of at least 20 out of 50. The second condition equals 40% of the exam. Treat the two tests separately when planning revision or calculating pass scenarios.
How do DFS, BFS, UCS and A* differ?
They choose the next frontier entry differently. DFS prefers depth through a stack, BFS prefers shallow paths through a queue, UCS prefers the lowest accumulated cost g, and A* prefers the lowest estimated total g+h. Their guarantees depend on cost and heuristic assumptions.
When is a heuristic admissible?
An estimate h is admissible when it never exceeds the true cheapest remaining cost to a goal. Zero is admissible but supplies no direction. A tighter optimistic estimate can reduce expansions while preserving the A* optimality condition taught in the course.
How does STRIPS update a state?
First check that every action precondition is in the current fact set. Then remove the delete effects, insert the add effects and preserve facts that the action does not mention. The plan succeeds only if the final state contains every required goal fact.
Where can I find past-paper practice?
The landed course materials include graph-search exercises, worked search solutions and STRIPS planning exercises, but they are tutorial resources rather than a past final paper. This guide therefore uses original questions shaped around expansion traces, heuristic checks, plan validation and modelling faults.
Questions students ask in each COMP9414 chapter
- Why is a convincing demonstration not enough?
- It establishes behaviour in the observed setting, but a broader claim also needs evidence about the mechanism, repeatability and task boundary. Changing wording, state or operating conditions can reveal whether the behaviour transfers or depends on a narrow setup. Chapter 01 · AI as Evidence and Capability
- What makes an agent rational?
- Rationality is relative to the stated objective, model and information. An agent can make the best available decision under uncertainty and still obtain a poor result. Evaluation should distinguish decision quality from luck and check whether the supplied utility matches the intended outcome. Chapter 02 · Rational Agents and Task Models
- What is the difference between a graph and a search tree?
- The graph represents each world state and its legal transitions. The search tree represents paths from the start, so the same world state can appear more than once under different parents. A tree node therefore carries plan history as well as a state label. Chapter 03 · States, Trees and Frontiers
- Why can BFS return an expensive route?
- BFS minimises the number of edges, not the sum of arbitrary weights. A two-edge route can cost more than a three-edge route. Its lowest-cost guarantee applies when every action has the same cost, so state the cost model before interpreting optimality. Chapter 04 · Search Strategy and Heuristic Choice
- What makes a planning action applicable?
- Every grounded precondition must be present in the current fact set under the tutorial's closed-world convention. If even one required fact is absent, the action cannot execute. Diagnosis should name the first illegal action and the specific false precondition. Chapter 05 · STRIPS Planning and Goal Interaction
How to study for the exam
Start with the exam hurdle and build reliability rather than collecting definitions. For graph search, redraw a small frontier after every expansion and record the path, depth and cumulative cost separately. Practise DFS and BFS on the same unweighted graph, then add weights and compare BFS with UCS.
For A*, keep a three-column g, h and f ledger and test admissibility against true remaining costs on graphs small enough to solve by hand. For planning, translate every action into preconditions, add effects and delete effects before reading a full plan. Stop at the first illegal action, name the false precondition, repair the order and replay the successor facts through to the goal.
Finish each session by explaining one guarantee with its assumptions and one model boundary that could cause execution failure. Use the official tutorial exercises for question type and this guide's independently authored examples for extra repetition. Check Moodle and your individual timetable for current final-exam logistics and permitted materials.
Your AI Computer Science tutor for COMP9414
Stuck on a hard COMP9414 question? Sia is AskSia’s AI Computer Science tutor — ask any COMP9414 Artificial Intelligence question and get a clear, step-by-step explanation grounded in how the course is actually taught and assessed. Read this whole study guide free, then take your hardest questions to Sia.
