Introduction to Artificial Intelligence - academic course

• lab_1.1 - Search Space vol. 1
• lab_1.2 - Search Space vol. 2
• lab_2 - Evolutionary and Genetic Algorithms
• lab_3 - Deterministic two-player games
• lab_4 - Regression and Classification
• lab_5 - Artificial neural networks
• lab_6 - Reinforcement learning (RL)
• lab_7 - Bayesian Models

Search Space - Knapsack problem

Task:

Find the optimal solution by exhaustive search. Solve the problem using a heuristic: pack items into a backpack in order of decreasing value-to-weight ratio.

Search Space - Stochastic Gradient Descent

Task:

Please implement the steepest descent/ascent method. We calculate the gradient numerically. Apply the method to find the minimum of the Booth function in 2 dimensions, then to find the minimum of functions 1 to 3 from CEC 2017 in 10 dimensions.

Evolutionary and Genetic Algorithms - evolutionary algorithm

Task:

Implement a classic evolutionary algorithm without crossover, using tournament selection and elitist succession. The available budget is 10000 evaluations of the objective function. We are optimizing functions number 4 and 5 from CEC 2017 in 10 dimensions. The bounds of the search space are -100 and 100.

Deterministic two-player games - MiniMax checkers

Task:

Implement the alpha-beta pruning min-max algorithm and apply it to the game of checkers/draughts. Let the evaluation function return the difference between the player's and the opponent's board state.

Regression and Classification - ID3 algorithm

Task:

Implement the ID3 classifier (decision tree) with nominal attributes and identity tests. Provide accuracy and confusion matrices for given datasets.

Artificial neural networks - two-layer perceptron

Task:

Implement a two-layer perceptron and train it to represent the function J: [-5,5] → R, given by the formula: J(x) = sin(xsqrt(5))+cos(xsqrt(3)).

Reinforcement learning (RL) - Qlearning agent

Task: Implement the Q-Learning algorithm and use it to determine a decision policy for the FrozenLake8x8 problem.

In addition to investigating the default reward system (1 for reaching the goal, 0 otherwise), please propose your own system of rewards and penalties, and then compare the results achieved with the default system.

Bayesian Models - Random data generator using Bayesian network distribution

Task:

Implement a random data generator that follows the distribution represented by a given Bayesian network. The network describes the dependencies between (binary) random variables and is provided as a graph structure and conditional probability tables in a text file. Divide the generated set and use it to train and test the classifier created in previous exercises.

git clone https://github.com/tilleyd/cec2017-py
cp -R cec2017-py/cec2017 .