COS 3751 (Techniques of AI) Analysis Based on Past Papers
Key Topics and Patterns
1. State Space Representation and Problem Solving
Consistently, each paper starts with questions on state space and problem representation
(e.g., representing puzzles and games in formal notation). Familiar examples include
puzzles like Francs and Pounds and scenarios like crossing a river.
Topics to focus on:
Differentiating between agent function and agent program.
Defining state representations for various puzzles.
Initial and goal states, action definitions, and applicable transitions.
2. Search Algorithms
All three exams include questions on search strategies such as Uniform Cost Search,
BreadthFirst Search, DepthFirst Search, and A*.
Exam expectations include:
Comparing search algorithms, often with specific attention to space complexity and order
of node expansion.
Applying A* search in grid or tree formats, showing steps and managing nodes with
given heuristic functions.
Recommended focus:
Practice with search trees and managing frontier lists.
Review heuristic functions and optimality conditions (e.g., admissibility for A*).
3. Adversarial Search and Game Theory
Each exam features adversarial search, typically involving Minimax or AlphaBeta pruning.
Questions often ask for specific values after running Minimax and checking if AlphaBeta
pruning is beneficial.
Key skills:
Calculating Minimax values and tracking AlphaBeta cuts.
Understanding scenarios where pruning reduces computation.
4. Constraint Satisfaction Problems (CSP)
, CSP questions appear in all papers, usually applied to realworld scenarios (e.g., shuttle
scheduling or simplified magic squares).
Areas to study:
Defining variables, domains, and constraints clearly.
Using constraint graphs to visualize dependencies.
Applying forwardchecking and domain consistency concepts.
5. Predicate Logic and Resolution
Each exam includes converting statements to First Order Logic (FOL), CNF, and
performing resolution refutation.
Core competencies:
Writing FOL representations of English statements.
Converting FOL to CNF and applying resolution for proof.
6. Machine Learning and Decision Trees
Decision tree questions are common, typically involving entropy, information gain, and
constructing trees from small datasets.
Focus points:
Calculating entropy and information gain.
Building decision trees based on calculated splits.
Addressing overfitting in decision trees and related strategies.
Key Topics and Patterns
1. State Space Representation and Problem Solving
Consistently, each paper starts with questions on state space and problem representation
(e.g., representing puzzles and games in formal notation). Familiar examples include
puzzles like Francs and Pounds and scenarios like crossing a river.
Topics to focus on:
Differentiating between agent function and agent program.
Defining state representations for various puzzles.
Initial and goal states, action definitions, and applicable transitions.
2. Search Algorithms
All three exams include questions on search strategies such as Uniform Cost Search,
BreadthFirst Search, DepthFirst Search, and A*.
Exam expectations include:
Comparing search algorithms, often with specific attention to space complexity and order
of node expansion.
Applying A* search in grid or tree formats, showing steps and managing nodes with
given heuristic functions.
Recommended focus:
Practice with search trees and managing frontier lists.
Review heuristic functions and optimality conditions (e.g., admissibility for A*).
3. Adversarial Search and Game Theory
Each exam features adversarial search, typically involving Minimax or AlphaBeta pruning.
Questions often ask for specific values after running Minimax and checking if AlphaBeta
pruning is beneficial.
Key skills:
Calculating Minimax values and tracking AlphaBeta cuts.
Understanding scenarios where pruning reduces computation.
4. Constraint Satisfaction Problems (CSP)
, CSP questions appear in all papers, usually applied to realworld scenarios (e.g., shuttle
scheduling or simplified magic squares).
Areas to study:
Defining variables, domains, and constraints clearly.
Using constraint graphs to visualize dependencies.
Applying forwardchecking and domain consistency concepts.
5. Predicate Logic and Resolution
Each exam includes converting statements to First Order Logic (FOL), CNF, and
performing resolution refutation.
Core competencies:
Writing FOL representations of English statements.
Converting FOL to CNF and applying resolution for proof.
6. Machine Learning and Decision Trees
Decision tree questions are common, typically involving entropy, information gain, and
constructing trees from small datasets.
Focus points:
Calculating entropy and information gain.
Building decision trees based on calculated splits.
Addressing overfitting in decision trees and related strategies.
