Useful for ASRB NET
Part of the machine-learning path: Machine Learning, Artificial Intelligence, Neural Networks and Deep Learning and Natural Language Processing.
This course is not about machine learning, and the difference is the point.
Machine Learning learns a function from examples. This course searches — it is given a description of a problem and finds a solution by exploring possibilities. Nothing here is trained on data.
| Machine learning (12 A) | Classical AI (this course) | |
|---|---|---|
| Given | Examples with answers | A description of the problem |
| Produces | A fitted model | A solution, or a path to one |
| Method | Optimise parameters | Search and inference |
| Knowledge | Implicit, in the weights | Explicit, in facts and rules |
| Explains itself | Poorly | Completely — you can print the chain |
| Fails by | Overfitting | Combinatorial explosion |
| Good at | Perception, prediction | Reasoning, planning, constraints |
Both are AI. This course is the older half — search, logic, knowledge representation — and it remains the right tool whenever the rules are known and the answer must be justified. A hospital triage system that must explain its reasoning is an expert system, not a neural network.
KEY INSIGHT
Units 2 and 3 are search. Given a state space, find a goal. Everything is a variation on which node do I expand next?
Units 4 and 5 are knowledge. Given facts and rules, derive new facts. Everything is a variation on what follows from what I know?
Unit 1 sets up the vocabulary for both.
| From | You have | Used here |
|---|---|---|
| Problem Solving Using C / 3 | Recursion, stacks, queues | Unit 2 is those data structures. DFS is a stack; BFS is a queue |
| Python Programming and Data Structures | Complexity, big-O | Every search strategy is compared on time and space |
| Statistical Foundations for Data Science | Probability, Bayes' theorem | Unit 5's Bayesian networks — Bayes again, on a graph |
| Machine Learning | Naive Bayes, evaluation | Unit 5 §5.4 is the same theorem; §5.6 contrasts the two paradigms |
| Database Management Systems | Relational algebra, queries | Prolog is a database that can reason. parent(X, Y) is a table |
| Computer Fundamentals and Office Automation | Boolean logic, truth tables | Unit 4's propositional logic is that, made into an inference system |
The lab is entirely in Prolog, a language you have not met. It is not like C, Python or JavaScript: you do not write how to compute, you write what is true and let the engine search.
Budget two weeks for the shift in thinking. §Lab setup below gets you running without installing anything.
Understand the fundamental concepts, history, types, and applications of Artificial Intelligence.
Develop problem-solving skills using state-space representations and search strategies for AI applications.
Apply informed and advanced search techniques including heuristics, local search, genetic algorithms, and constraint satisfaction problems.
Learn knowledge representation methods and reasoning techniques using propositional and first-order logic for intelligent agents.
Explore expert systems, probabilistic reasoning, fuzzy logic, and emerging AI technologies including NLP, robotics, and ethical considerations.
Definition and scope of AI; history, the two winters and the Turing Test; applications; Weak against Strong AI and Narrow against General AI; the structure of an agent and what rationality actually requires; the five agent types; PEAS; and the six environment properties — observable, deterministic, episodic, static, discrete and single-agent.
UNIT 2State space representation and its four components; problem formulation for the 8-puzzle, water jug and vacuum world; the general search algorithm and the frontier; completeness, optimality, time and space complexity; Breadth First, Depth First and Uniform Cost Search measured against each other on the Romania map; iterative deepening.
UNIT 3Heuristics, admissibility and consistency; Greedy Best First Search and why it is short-sighted; A*, its optimality proof and what an inadmissible heuristic costs; heuristic dominance; local search, hill climbing and its local maxima; simulated annealing and the cooling schedule; genetic algorithms; constraint satisfaction, backtracking, MRV, degree and LCV.
UNIT 4Knowledge-based agents and the TELL/ASK interface; representation issues and the four approaches; propositional logic syntax, semantics, truth tables, validity, satisfiability, entailment and inference rules; first order logic, quantifiers, substitution and unification; forward and backward chaining and when each is right; resolution, CNF and proof by refutation.
UNIT 5Expert system architecture — knowledge base, working memory, inference engine and explanation facility — and why the explanation falls out of the proof; Bayes theorem and Bayesian belief networks; fuzzy logic and how degree of truth differs from probability; NLP basics; robotics; AI ethics, bias, accountability and societal impact.
PRACTICEExam-style questions with fully worked solutions.
LABEvery prescribed lab experiment, with code and expected output.