UROP Openings

Have a UROP opening you would like to submit?

Please fill out the form.

Submit your UROP opening

What Do You Mean? Learning to Play Cooperative Games with Limited Communication


Term:

Fall

Department:

6: Electrical Engineering and Computer Science

Faculty Supervisor:

Gregory Wornell

Faculty email:

gww@mit.edu

Apply by:

September 23 2020

Contact:

Professor Gregory Wornell, Gary Lee and Tejas Jayashankar: gww@mit.edu, garyleecf@mit.edu, tejasj@mit.edu

Project Description

Many games and puzzles are simplified versions of real-life situations or engineering problems. Indeed, learning how to solve them can therefore have surprising practical impact. This is also why games play a critical role in artificial intelligence (AI) research. Recently, we have seen computers equipped with AI achieving better-than-human performance in competitive games like Chess and Go, where typically one player wins and another loses. However, a new frontier in AI development are games requiring cooperation with other “agents” (humans or other machines) to achieve a common goal. Typically, communication between the agents is limited (e.g., in how often they can communicate and what they can say). In such settings, an intelligent agent needs to infer the knowledge, beliefs, and intentions of other agents from their constrained behaviors. We will use both the card game of Hanabi and inductive hat puzzles as sandboxes to study cooperation under imperfect information and limited communication. In these settings, other agents possess information that you may not have, and we seek good strategies within the rules for each agent to convey what they know. In this project, you will be investigating and applying AI tools. Starting from a progression of simple simulation environments, you will explore ways a computer can learn to perform well. -- A team of three students will collaborate on this project. The goals and tasks of the project will be tailored to accommodate the experience and interests of the team. (First-year students are particularly welcome and encouraged to apply!) Each student appointment is for 10 hours per week over 14 weeks in the Fall semester (and potentially into IAP). All project meetings will be via zoom. The hourly wage for the position is $13/hour for a maximum of 140 hours. Students must be enrolled for the Fall semester and currently residing in the US to be eligible (though a credit-only UROP position for those residing outside of the US may be possible.) To apply, send a short statement of interest and CV/resume to Professor Gregory Wornell, Gary Lee and Tejas Jayashankar (gww@mit.edu, garyleecf@mit.edu, tejasj@mit.edu). We are separately advertising a second, complementary ELO experience, “To Send or Not to Send? : Learning to Share the Airwaves in Wireless Networks.” If you are also applying for this ELO, let us know your preference between the two (if any). Application deadline is September 23, 2020, but applications will be considered on a rolling basis. Eligible applicants will be invited to an interview. You may expect to hear back not later than September 30, 2020. If you have any questions about the ELO, feel free to contact us by email.

Pre-requisites

Prospective members of the team should be comfortable using basic computational tools. Experience with programming, particularly in Python, is preferred but not required; the project provides opportunities to learn and improve your programming skills.