📄️ Foundation Models
Discussion on T0pp. ChatGPT was released later. You can reduce a lot of downstream tasks into prompting. Big deal in 2022 or thereabouts.
📄️ Health Data Modalities I
Berkson's Paradox. A and B are independent. Both cause C. Now A and B appear related.
📄️ LLMs
Discussion of Chinese Room Paradox.
📄️ Health Data Modalities II
How do the data differences manifest in modeling choices?
📄️ Health Data Modalities III
The text is the richest but also the messiest and rapidly changing part of EHR. Byproduct of clinical care and billing. See it as a series of nominal tokens.
📄️ Modalities - Imaging
Images are the first 'grid' data type (EHR is irregular sequences) and are quite amenable to the kinds of representaions ML needs. Lots of great work here. Lots of modalities.
📄️ Imaging Modeling
Transfer learning: Kind of the source of the first major "AI Summer" or DL becoming dominant in AI in general. Driven by ImageNET. Leverage the early layers of that network and fine-tune for downstream tasks.
📄️ Causal Inference
Prediction versus Intervention. Former: "Among patients who got the treatement, what is $Y$?"
📄️ DNA, Genetics, and Gene Regulation
Discussion on Central Dogma. DNA -> RNA -> Protein. Every step has intermediate regulatory processes.
📄️ Proteins, Molecules, & Structural Biology
64 codons -> 20 amino acids + stop signals (UGA, UAA, UAG). Always 5' -> 3' end. Sequence constrains Structure. Structure is inextricably linked to function.
📄️ Modern Biological AI
Rapid advances in this space rivalled only by NLP advances. This is about what deep learning adds.
📄️ Random
Lecture 10 2026-02-19 0919