6.S198 Deep Learning Practicum


Level: Undergraduate
Units: 3-3-6
Prerequisite: 6.034 or 6.036, or permission of instructor
Instructor:  Prof. Hal Abelson (hal@mit.edu)
Schedule: MW2-3:30, room 36-112
This subject counts as a department laboratory subject.
This course prepares students to carry out projects that use deep learning.  In the first part of the course,  we’ll survey basic techniques, including convolutional neural networks, recurrent neural networks, generative adversarial networks, and embedding.  For each technique, we’ll begin by studying the fundamental concepts, explore open-ended demo applications that use the tool and carry out programming assignments to modify and extend the demo.  For the second part of the course, students will work in teams to design and implement original projects that use these techniques.  Each project team will have a mentor who is a machine learning expert from industry.  We’ll also discuss some policy and social related issues in deep learning, such as algorithmic fairness  and interpretability.

Programming exercises will be done in Javascript and Typescript (a variant of Javascript), using a new open source platform called deeplearn.js, which is a simplified version of Tensorflow that runs in the browser.

6.S198 counts as a course 6 advanced departmental lab (DLAB2).

Suggested  prerequisites are a working knowledge of Javascript (e.g., as in 6.170) and some exposure to machine learning (e.g., as in 6.034 or 6.036) or permission of instructor.

Enrollment is limited, permission of instructor required.  To apply for permission to enroll in 6.S198 for the spring, you fill out the web form at: https://mit.edu/6.s198.