Catalog Description
(New topic in development for fall 2027 or spring 2028)
Survey modern generative AI models, including large language models and diffusion models, and the methods used to apply them in practice. Weekly lectures, paper discussions, and programming assignments build toward a final capstone project.
Instructor Description
This course offers a solid introduction to modern generative AI. It opens with the foundations: probabilistic modeling, the major families of generative models, and the neural architectures behind them, from CNNs and RNNs to transformers. It then examines how large language models are pretrained and scaled, and how diffusion models generate images through iterative denoising. The middle of the course focuses on practical methods: adapting models with fine-tuning and LoRA, aligning them through preference optimization, and steering them with prompting and reasoning strategies. Participants also build retrieval-augmented and tool-using systems and learn to evaluate and serve them efficiently. The final part covers real-world applications in software, science, healthcare, and creative work, along with their risks, including bias, privacy, copyright, and safety. Each week pairs a lecture and a paper discussion with a programming assignment, so every technique is practiced as it is introduced. Later in the semester, participants present and critique a research paper. The course closes with final project presentations, including a candid analysis of each project’s failure modes.
Prerequisites
Graduate Standing. Introductory-level familiarity with Python programming recommended.
Restrictions
Enrollment in Information Studies (INF) courses is restricted to graduate students in the School of Information through registration periods 1 and 2, with outside students only being accepted during period 3.