Course positioning
Taught by Serena Yeung, CS 286 discusses visual foundation models in biomedicine in depth. It covers representation learning, diffusion/generative models, and vision-language models, as well as real deployment issues such as model size, compute, training/inference settings, and data.
What it covers
- Visual FMs: CLIP, DINO, SAM, etc.
- Representation learning: self-supervision, contrastive learning
- Generative models: diffusion, image translation
- Vision-language: report generation, VQA
- Compute and data: model scale, training/inference, data bias
- Medical applications: pathology, radiology, surgery
- Paper analysis and project
Prerequisites
CS 231N or BIODS 220-level deep learning/vision background.
Assessment & workload
Paper analysis plus class discussion and a final project. Check the current syllabus for details.
Who it's for
Graduate students in medical imaging, CV research, or AI4Science.
Study advice
Take CS 231N first. The course is research-oriented; pick a narrow clinical problem and go deep.
Related courses
Official course description
This course in artificial intelligence will provide a deep dive into advanced computer vision techniques for reasoning about visual data, and their real-world use in biomedicine. We will cover current cutting-edge models including different families of vision foundation models from representation learners to diffusion and generative models, and both vision-only and vision-language models. We will also cover considerations for real-world use, including model size and computation, training and inference settings, and training data, focusing on applications in biomedicine. Students will be actively engaged in studying and analyzing recent advancements through written analyses, class discussions, and a final project. This course is considered an advanced course and students should be comfortable with deep learning and computer vision at the level of CS231N or BIODS220.