(HL) Unsupervised, reinforcement and evolutionary learning, and neural networks

Learning objectives

(HL) Describe how clustering algorithms such as k-means group unlabelled data by similarity of features, for example to segment customers from their purchases

(HL) Describe how association rule learning finds attributes that occur together in large data sets, and interpret the support and confidence of a discovered rule

(HL) Describe how a reinforcement-learning agent takes actions in the states of its environment to maximise cumulative reward, balancing exploration against exploitation

(HL) Describe how a genetic algorithm improves a population of candidate solutions through a fitness function, selection, crossover, mutation and a termination condition, for problems such as the travelling salesperson

(HL) Outline the inputs, weights, bias, activation function and output of a perceptron, and explain how the hidden layers of a multi-layer network model complex patterns

(HL) Describe how a convolutional neural network classifies images through convolutional layers with kernels and a stride, activation functions, pooling layers and fully connected layers, and how the number of layers, kernel size, stride, activation and loss function affect performance