Lecture
This lecture introduces Gaussian Mixture Regression (GMR), a method to estimate joint and conditional densities from multimodal datasets. It covers the principles of GMR with single and multiple Gaussian components, explaining how to estimate marginals, conditionals, and regressions. The lecture also discusses the computation of variances and the interpretation of noise in state space. The key takeaway is that GMR allows for predicting multi-dimensional outputs and capturing correlations across dimensions, although it requires computing the full distribution and careful initialization of the Expectation-Maximization algorithm.