Course positioning
STATS 315A is a graduate-level ML course in the statistics department, covering prediction methods (linear models, trees/forests, deep networks), model evaluation (cross-validation, calibration, conformal prediction), optimization (convexity, stochastic methods), and generative models (language models, diffusion, VAEs). It balances statistical rigor with modern methods.
What it covers
- Prediction methods: linear, trees/forests, deep networks
- Model evaluation: CV, calibration, conformal prediction
- Optimization: convex optimization, SGD, adaptive methods
- Generative models: language models, diffusion, VAEs
- High dimensions and regularization: lasso, sparsity
- Project/paper discussion
Prerequisites
Probability/statistics, linear algebra, ML basics (STATS 202/CS 229 level).
Assessment & workload
Assignments plus a project/exam. Check the current syllabus for details.
Who it's for
Statistics/ML graduate students; people who want rigorous predictive modeling and uncertainty quantification.
Study advice
Take STATS 202 or CS 229 first. The conformal prediction part is worth going deep on.
Related courses
Official course description
This course covers statistical techniques underlying modern machine learning. Topics include prediction methods (linear models, trees and forests, deep neural networks), model evaluation (cross-validation, calibration, conformal prediction), optimization (convexity, stochastic methods, adaptive metrics), and generative models (language models, diffusion models, variational autoencoders).