Principles of neural computation in brains and machines

Course numbers

9.S52 (U; undergrad) / 9.S912 (G; grad)

Instructors

Professor: Guangyu Robert Yang, PhD (yanggr@mit.edu)

TA: Quilee Simeon (qsimeon@mit.edu)

Meeting times

Lectures: 2 sessions / week, 90 minutes / session, MW 9:30am-11am

Classroom

Building 46 Room 1015 (46-1015)

McGovern Institute for Brain Research / BCS building, 1st floor level (entrance is on Vassar Street)

Prerequisites

Required

18.C06 18.06 Linear Algebra, 6.0001 Intro to CS & Programming, 9.01 Intro to Neuroscience, Experience with Python programming, Basics of Pytorch and Machine learning

Recommended

Experience with deep learning and/or computational neuroscience

Course description

Brains and many machines rely on computation in neural networks. This course introduces and discusses some of the core principles behind neural computation in brains, machines, or both. Here by principle, we refer loosely to big ideas that we choose to keep if we must lose all other knowledge.

The course has two components. Each component consists of a number of modules.

The first component is Views, which describes different ways we can think about neural computation. This component consists of Representation, Algorithm, and Dynamics modules.

The second component is Principles, which covers some basic principles of neural computation. Each module is structured in the following way:

What is this principle?
What is an example of this principle applied?
Why is this principle fundamentally important? Intuition and/or Theory
What are some other examples of this principle in brains and machines?
(Optional) What is currently missing from our understanding?

There will be lots of open discussions during the lectures.