BINF GU 4002: Machine Learning for Healthcare

Staff:

Instructor: Shalmali Joshi
Teaching Assistants: Young Sang Choi and Yuta Kobayashi

Logistics:

Contact: Courseworks
Location: 622 W 168th St, PH20-200, NY-10032 on the 20th floor of the Presybterian Hospital Building (Directions)
Lecture Times: Tuesdays, Thursdays 9:00a - 10:15am
Instructor Office Hours: Shalmali Joshi: Fridays TBD at 622 W 168th St, PH-20, 402
TA Recitation and Office Hours (optional): Young Sang Choi and Yuta Kobayashi: Fridays TBD, at 622 W 168th St, PH-20, Collaboration Room

Pre-reqs:

Basic knowledge of probability, statistics, calculus, and linear algebra is expected and required.
Familiarity with programming with Python and versioning systems like Git is required.

Course Description:

Machine Learning (ML) has transformative potential for applications in health and medicine. You will learn fundamentals of machine learning, deep learning, and modeling clinical and biomedicine data using machine learning.

Objective:
By the end of this course you will be able to:

Learn foundations of major machine learning concepts and their relevance to health and biomedicine data
Develop new ML methods focused on health/biomedicine tasks
Be better equipped to evaluate and validate of ML in healthcare and medicine

Course Material:

Course material will largely consist of lecture notes, and specific references provided for each module. No textbook is required, though some textbook chapters may be provided as additional reading.

Policy on Generative AI:

Use of generative AI for understanding concepts and reading material is allowed.
We will not allow the use of generative AI for solving homeworks or doing machine learning modeling. Homework reports should not be generated using any AI tool. You will be given 0 points for the complete homework assignment if any violations are detected.

Schedule and Content

Note that schedule and content is subject to minor changes. Slides and notes will be posted on Courseworks.

Lecture number	Date	Day	Topic	Subtopics	Homework timelines
1	2024-01-21	Tuesday	Intro and course logistics, MIMIC-IV access, Google Colab Credits	Colab accounts setup, data downloads, conda environment setup, etc.	Homework 0 out with instructions on setting up Colab, accessing MIMIC-IV data
2	2024-01-23	Thursday	Probability and information theory primer		Homework 0 due, Homework 1 out
3	2024-01-28	Tuesday	Linear algebra and optimization primer
4	2024-01-30	Thursday	Introduction to supervised learning	Empirical Risk Minimization, Loss functions, Model families
5	2024-02-04	Tuesday	Introduction to supervised learning	IID vs OOD, Bias-variance tradeoff, regularization
6	2024-02-06	Thursday	Empirical practices in machine learning	LOOCV, validation data, calibration, uncertainty quantification, bootstrapping
7	2024-02-11	Tuesday	Principles of Maximum Likelihood	Logistic regression- two views, other models motivated by maximum likelihood estimation
8	2024-02-13	Thursday	Basics of probabilistic modeling and Bayesian inference	Prior, Likelihood, and Posterior. Logistic and Linear Regression
9	2024-02-18	Tuesday	Basics of probabilistic modeling and Bayesian inference	Posterior Predictive, Exponential Familities, Maximum-a-Posteriori	Homework 1 due, Homework 2 out
10	2024-02-20	Thursday	Introduction to Regression	Linear regression, other types of regression
11	2024-02-25	Tuesday	Bayesian linear regression	Derivation and connection to regularization
12	2024-02-27	Thursday	Empirical practices in machine learning - revisited	Comparing approaches to uncertainty quantification, best practices, etc.
13	2024-03-04	Tuesday	Review of basic supervised learning	Decision trees, Random Forests, XGBoost
14	2024-03-06	Thursday	Introduction to deep neural networks	Multilayer perceptron and connection to logistic and linear regression
15	2024-03-11	Tuesday	Optimization in deep neural networks	Backpropagation, stochastic gradient descent	Homework 2 due, Homework 3 out
16	2024-03-13	Thursday	Midterm	Midterm
No class	2024-03-18	Tuesday	Spring break	Spring break
No class	2024-03-20	Thursday	Spring break	Spring break
17	2024-03-25	Tuesday	Deep learning for image data	Convolutional Neural Networks
18	2024-03-27	Thursday	Deep learning for sequential data	Recurrent Neural Networks, LSTM, State-space models, Gated Recurrent Units
19	2024-04-01	Tuesday	Deep learning for networked data	Graph Neural Networks
20	2024-04-03	Thursday	Deep learning for sequential data	Transformer: Attention-based neural networks
21	2024-04-08	Tuesday	Deep learning for sequential data - contd	Training paradigms for sequence based models (e.g., Seq-2-seq, decoder-only etc)	Homework 3 due, Homework 4 out
22	2024-04-10	Thursday	Distribution shifts, generalization, and domain adaptation	Concept of generalization, types of distribution shifts, examples of implications in healthcare
23	2024-04-15	Tuesday	Distribution shifts, generalization, and domain adaptation - contd.	Focus on various methods of adaptation to overcome different types of distribution shifts
24	2024-04-17	Thursday	Unsupervised learning	History and review of classical methods, brief review of modern methods
25	2024-04-22	Tuesday	Generative modeling	Foundations of generative model, basic loss-functions and a broad overview of models
26	2024-04-24	Thursday	Foundation models -LLMs	Large-language models (Transformers but more)
27	2024-04-29	Tuesday	Foundation models - Vision-language	CLIP and other basic Vision-language models	Homework 4 due
28	2024-05-01	Thursday	Foundation models- Biological data - e.g., AlphaFold	Major foundation models for biological data
29	2024-05-05	Monday	Finals

Format and Grading

Lectures held twice a week (5 points for participation in-person)

Homeworks:Problem Sets: Each homework will be a combination of numeric/conceptual/computational problem and data-driven/modeling questions (4 homeworks, 40 points). A 'Homework 0' will be assigned in the first week on basics of probability and linear algebra.

Homework Assignments (40 points)

Homeworks will be assigned on Courseworks. Homeworks will be due 3 weeks after they are assigned. We have ensured ample time for all homeworks. However, we are allowing two slack days usable over the semester. If you are using a slack day for a homework, mention this on your homework submission (otherwise the TAs will assume late submission). You are free to use the slack days for any of the homeworks.

Exams

There will be one mid-term (25 points) and one final exam (30 points). Both exams will be for the duration of the class. You will be allowed one Letter-sized cheat sheet for formulae only. You will return the cheat sheet along with the exam for grading.