PSI 2018/2019 - Machine Learning (Hayward)

Motivation; goals of machine learning and many-body physics; definitions of supervised, unsupervised and reinforcement learning

Supervised learning with feedforward neural networks: weights and biases, activation functions, cost functions, gradient descent

Alternatives to gradient descent, backpropagation, recognizing overfitting

Monte Carlo sampling; summary of hyperparameters in feedforward neural networks

Convolutional neural networks: local receptive fields, shared weights and biases, pooling

Introduction to unsupervised learning; dimensional reduction using principal component analysis

Dimensional reduction using t-distributed stochastic neighbour embedding (t-SNE); Kullback-Liebler (KL) divergence; maximum likelihood estimation

Reinforcement learning: Markov decision processes, policy gradient methods

Reinforcement learning: Q-learning, Bellman equations

Generative modelling: Hopfield networks, Boltzmann machines and restricted Boltzmann machines (RBMs)

Restricted Boltzmann machines: training to minimize the Kullback-Liebler divergence