PHYS-467: Machine learning for physicists

Lectures in this course (106)

Explores the statistical physics of clusters, focusing on complexity and equilibrium behavior.

Explores modern system internals and how they solve complex tasks.

Discusses the unified formulation of the generalized linear model and the optimization of loss functions.

Explores Biased Monte Carlo Markov Chain, including Bayes-optimal estimation and Metropolis-Hastings algorithm.

Covers the Expectation Maximization algorithm for learning parameters and dealing with unknown variables.

Explores hidden variables, graphical models, and computational gaps in neural network learning.

Covers generative neural networks, focusing on sampling, training, and noise addition.

Covers PCA application and BBP phase transition in a card game dataset.

Covers the implementation of Markov Chain Monte Carlo (MCMC) with the Metropolis algorithm for sampling from posterior distributions.

Covers the K-means algorithm for clustering data samples into k classes without labels, aiming to minimize the loss function.

Covers the Boltzmann Machine, a type of stochastic recurrent neural network.

Covers Bayes estimator, Simulated Annealing, and EM for parameter estimation.

Covers the concept of kernel regression and K-nearest neighbors for making data linearly separable.

Explores kernel regression, the curse of dimensionality, and random features in neural networks.

Covers the implementation and analysis of the K-Means algorithm, including centroid initialization, point assignment, and noise impact.

Covers loading datasets, understanding dimensions, learning curves, and the impact of regularization on overfitting.

Explores random features in neural networks and kernel regression using stochastic gradient descent.

Covers the fundamentals of multi-layer neural networks and the training process of fully connected networks with hidden layers.

Covers feature selection, kernel regression, and neural networks through exercises.

Explains back-propagation in neural networks, updating weights based on errors and evaluating networks through training and test losses.

Page 5 of 6