Skip to main content

CS 9840: Probabilistic Generative AI

Summarized course website for CS 9840: Probabilistic GenAI, Winter 2026. Students will learn the theoretical foundations for generative AI.

From chatbots to image generators, Generative AI (GenAI) is reshaping the world. What algorithms and probabilistic principles power this revolution? This course introduces the theoretical and algorithmic foundations of GenAI, covering large language models (LLMs) and diffusion models, among more mature methods. We will also explore applications in AI for Science, such as drug discovery and weather forecasting.

info

Detailed course information is on OWL Brightspace (available starting December 2025) and the course outline.

Prerequisites

There are no formal prerequisites. But, since GenAI is inherently probabilistic, some knowledge of the following is necessary.

tip

There will be review sessions (Weeks 1 & 2). However, students who are not comfortable with these concepts are highly encouraged to independently read the D2L book, Chapter 2.

  • Probability/statistics:
    • sum, product, and Bayes' rule,
    • probability distributions (Gaussians, Categorical),
    • maximum likelihood.
  • Linear algebra:
    • vectors and matrices,
    • matrices/vectors products,
    • matrix inversion.
  • Multivariable calculus:
    • partial derivatives, gradients, Jacobians, & Hessians,
    • multivariable integration.
  • Machine learning:
    • regression, classification,
    • neural networks,
    • gradient descent,
    • familiarity with deep learning libraries (code examples are in PyTorch).

Schedule

info
  • Lecture: Tuesdays, 10 AM–12 PM @ MC 320.
  • Tutorial: Thursdays, 12:30 PM–1:30 PM @ MC 320.
  • Office hour: Thursdays, 1:30 PM–3:00 PM @ MC 362.

WeekDateLecture (Tue)Tutorial (Thu)Resources
1Jan 6Introduction & review of fundamentalsReview of fundamentalsBishop 2024, Chapters 2 & 3; D2L, Chapter 2
2Jan 13Variational autoencodersReview of fundamentalsBishop 2024, Chapter 19; D2L, Chapter 2
3Jan 20Explicit and implicit modelsHW 1Bishop 2024, Chapters 17 & 18
4Jan 27Application: De novo drug discoveryHW 2Gómez-Bombarelli, et al., 2018
5Feb 3Diffusion Models I: IntroductionProject info sessionBishop 2024, Chapter 20
6Feb 10Diffusion Models II: Advanced TopicsProject proposal deadline & project Q&ABishop 2024, Chapter 20
7Feb 17Reading weekNo classes
8Feb 24Application: Climate modelingHW 3
9Mar 3LLMs I: ArchitecturesStylistic scientific writing 101Bishop 2024, Chapter 12; D2L Chapter 11
10Mar 10LLMs II: Decoding and reasoningHW 4Bishop 2024, Sec. 12.3.2; D2L Sec. 10.8; Lightman et al., 2023
11Mar 17LLMs III: Finetuning and alignmentSelected topics in GenAIBishop 2024, Chapter 12; Steinnon et al, NeurIPS 2020; Hu et al., ICLR 2022
12Mar 24Project presentationsProject presentations
13Mar 31Project presentationsProject presentations

info
  • HW 1-4 must be submitted the midnight before the corresponding tutorial. For example, HW 1 must be submitted by Wednesday, Jan 21, 11:59 PM.
  • Project proposal must be submitted by Wednesday, Feb 12, 11:59 PM.
  • Project presentation slides must be submitted by Monday, Mar 23, 11:59 PM. The presentation order will be drawn randomly.
  • Project final report must be submitted by Thursday, Apr 9, 11:59 PM.

References

  1. Bishop, "Deep Learning: Foundations and Concepts", Springer. 2024. [Link]
  2. Zhang, Lipton, Li, and Smola, "Dive into Deep Learning (D2L)". [Link]
  3. Gómez-Bombarelli, et al., "Automatic chemical design using a data-driven continuous representation of molecules." ACS Central Science 4(2). 2018. [Link]
  4. Price, et al., "Probabilistic weather forecasting with machine learning", Nature 637(8044), 2025. [Link]
  5. Stiennon, et al., "Learning to summarize with human feedback", NeurIPS, 2020. [Link]
  6. Lightman, et al., "Let's verify step by step." ICLR. 2023. [Link]
  7. Hu, et al., "LoRA: Low-rank adaptation of large language models", ICLR, 2022. [Link]

Grading

Students will be asked to form groups of 2 in the early weeks of the course. All homeworks and project are to be done in group. Homework assignment sheets will be available on OWL Brightspace. Refer to the course outline for detailed information.

TypeWeightCountTotal
Homework10%440%
Project Proposal15%115%
Project Final Presentation20%120%
Project Final Report25%125%
Total--100%

Policies

  • GenAI such as LLM chatbots must not be used to generate HW solutions and reports/slides.
  • Acceptable use of GenAI: (Must be disclosed as at the end of each document.)
  • Late submissions will not be marked since HWs and project are done in a group.

Refer to the course outline for more info and other polices.