Winter 2026: STAD68 Advanced ML
Summer 2021, 2022: Deep Learning Theory
Winter 2021: STAC58 Statistical Inference
Winter 2020, 2021, 2022, 2023, 2026: STAD80 Analysis of Big Data
Fall 2019: STA4527 Random Matrix Theory and Its Applications
Summer 2019: High Dimensional Statistics
Winter 2018, 2019, 2020: STAC63 Probability Models
Winter 2018, 2019: STA3000 Advanced Theory of Statistics
The prerequisites for STAD80 are STAC58H3, STAC67H3, and CSCC11H3. Please plan ahead if you want to take this course.
There won't be textbooks for this course.
This course's materials evolve from year to year.
This course has a heavy workload.
This course uses independent research projects to evaluate the student performance in part.
All works must be typed, using Latex or jupyer notebook, and submitted at Quercus.
No late work.
This course is by no means a bird course. If you want an easier one, take something else.
Do not cheat in this class, just drop it and take an easier one. No one cares whether you took this course or not.
Introduction
ANN
CNN
Sequence Models
Transformers
Mamba
Training your LLMs
Post-training
Agentic AI
Other potential topics include: Flow and diffusion models, deep ensembles, model merging, and model eidtting, if time allows.
The prerequisites for STAD80 are STAC58H3, STAC67H3, and CSCC11H3. Please plan ahead if you want to take this course.
There won't be textbooks for this course. However, we will develope a set of lecture notes along the way.
This course's materials evolve from year to year.
This course has a heavy workload.
This course has a midterm and uses an independent research project to evaluate the student performance in part.
All works must be typed, using Latex, and submitted at Quercus.
No late work.
This course is by no means a bird course. If you want an easier one, take something else.
Do not cheat in this class, just drop it and take an easier one. No one cares whether you took this course or not.
Fundalmentals
Predictive modeling
From predictive to generative modeling
Flow and diffusion models
Training flow and diffusion models
Controllable generation
Discrete flow matching and diffusion language models
Sampling faster: Advanced Solvers
Sampling faster: Distillation
Sampling faster: Training from scratch
Other potential topics include: Riemannian flow models, Muon, applications such as inverse problems, alignment/postraining, if time allows.
I will have a new semester-long course on Trustworthy AI in the year of 2023--2024. Stay tuned and check back later.
Unfortunately, I no longer have time to supervise undergraduate independent research projects except in exceptional circumstances. If you believe your background is particularly strong—for example, if you have published in top-tier ML/Stats venues, IMO medalist, or are an ICPC Finalist — please feel free to reach out. Otherwise, I encourage you to consider taking my courses, such as D68 and D80, instead.
Understanding Deep Learning by Simon Prince
Machine Learning 2021 by Hung-yi Lee (focus on general deep learning)
Machine Learning 2023 by Hung-yi Lee (focus on generative AI)
Generative AI 2024 by Hung-yi Lee
Transformers at Hugginface
Trustworthy ML by Kush R. Varshney (focus on concepts)
Trustworthy ML by Mucsányi el al. (focus on OOD generalization, XAI, and uncertainty quantification)