COURSE DESCRIPTION:
This elective course will offer a practical introduction to modern methods in Artificial Intelligence (AI) for biomedical research, by means of hands-on coding workshops and participation in a coding challenge.
During the last decade, AI has achieved impressive progress in real-world applications that appeared out of reach just a few years ago, and now features regularly in both scientific and general news. Modern AI methods are also becoming increasingly recognized as an integral component of the toolbox of the biomedical research community. Students will be grouped in one or more teams (depending on the number of students enrolled). Each team will design and implement a model, based on the resources learning in the course, to attempt to solve the specific challenge proposed in the competition. An additional goal of participating to the competition, is to introduce students to scientific teamwork: they will be guided to divide the challenge into smaller tasks (e.g. data preprocessing, designing the architecture, model training and evaluation), and work in synergy.
The practical resources and experience gained through this course will provide an entry point for students that may be interested in further understanding those AI tools in depth (i.e. to become developers) or in using them in practice in their thesis research (i.e. to become informed users).
The course will last seven weeks, with one class per week. During the first three weeks, students will learn the PyTorch framework: how to implement and apply key algorithms and architectures (e.g. CNN, RNN, Transformer) and how to work with datasets for training and evaluation. In addition, details of the competition will be explained. During the next four weeks, teams will work on their design and implementation.
COURSE OBJECTIVES:
To learn how to use foundational techniques in machine learning and modern AI methods, and apply them to a specific biomedical research question, through participation in coding workshops and developing solutions to a coding challenge.
To provide a practical introduction to software and computational tools that support the most important paradigms in modern AI, through hands-on coding tutorials and workshops.
To demonstrate the application of these tools to a real-world biomedical research question, through participation in an open challenge or competition (e.g. brain-score.org and algonauts.csail.mit.edu, to predict neural activity patterns in response to visual stimuli).
PREREQUISITES:
Recommended course: Introduction to the Mathematics of Theoretical Systems Biology (Block I).
Familiarity with Coding in Python.
It is also recommended that students be familiar with the following maths, prior to enrolling in this course:
- Linear algebra (matrix-vector operations; tensors).
- Calculus (integrals, derivatives and function optimization).
- Probability and statistics (Marginal/conditional/joint distribution; Bayes rule; common distributions e.g. Gaussian, Gamma, Bernoulli, Poisson).
Students who need a refresher with 1-3 can use this online resource
https://compneuro.neuromatch.io/tutorials/intro.html in the ‘pre-reqs refresher’ section for introductory tutorials.
- Foundational concepts of machine learning and deep learning (supervised and unsupervised learning; multilayer perceptron, convolutional neural networks; cost function, backpropagation)
Students not familiar with 4 should take Module 1 (Foundations) of this course concurrently with Module 2 (Coding).
REQUIRED MATERIALS:
Laptop with an internet browser for tutorial classes. Students should create an account on Google Colab.
SUITABLE FOR 1ST YEAR STUDENTS:
Yes.
STUDENT ASSESSMENT:
During the workshops on the first three weeks, students will be required to complete coding exercises. During the next four weeks, students will work on their assignment (a submission for the coding challenge).
Grades will be based on completing the exercises in class (or as homework, if needed; 50%), implementing a working model for the challenge (25%), and final presentation of the model design and results (25%).
Individual grades for the model will be identical for all students in the group. For the final presentation, each student will present a different aspect of the work, and individual grades will be based on the clarity and depth of presentation of each student’s portion of the presentation. The final grade (H/P/F) is cumulative and will be transmitted at the end of the course. A grade of H will be obtained by students whose model for the challenge performs competitively with respect to basic benchmarks (as provided by the organizers of the challenge) and whose presentation is delivered clearly, in addition to completing all exercises. A student will receive F if they consistently fail to complete exercises.
COURSE ATTENDANCE POLICY:
Students are expected to attend all classes. If missing a workshop class, students will have to complete the exercises on their own time (code and guidance will be provided).
CREDIT HOURS: 1.0
