Gaussian-Mixture Classification

Per-task GMM with K components → K-class labels

Random Gaussian mixtures in D dimensions. Trains a PFN that does Bayesian-optimal classification on any well-separated mixture.