UROP Openings

Have a UROP opening you would like to submit?

Please fill out the form.

Submit your UROP opening

Continual Acquisition of Visual Concepts


Term:

Fall

Department:

6: Electrical Engineering and Computer Science

Faculty Supervisor:

Antonio Torralba

Faculty email:

torralba@mit.edu

Apply by:

Anytime

Contact:

Shuang Li: lishuang@mit.edu

Project Description

Humans are able to rapidly learn and integrate skills with prior knowledge. In contrast, current deep learning models do not exhibit this basic skill. We study this continual learning problem in the image generation setting. This project aims to continually learn new object attributes and generate new combinations. For example, given an image of a “flying bird” and another image of a “running horse”, human has the ability to imagine an image of “flying horse”. Our model has a similar imagination/generation ability. Please refer to our previous paper https://arxiv.org/pdf/2004.06030.pdf for more information.

Pre-requisites

Experience with Pytorch or Tensorflow. Have enough time on the project every week