Audrey Cui

Hello! I'm Audrey, a rising sophomore studying course 6-9 (Computer Science & Cognitive Science) at MIT. As an aspiring researcher and avid artist, I'm interested in the ways computers can artistically augment and help us understand human creativity.

As an undergraduate research assistant at MIT CSAIL, I'm working on generative adversarial network (GAN) inversion, feature representation, and controllability. This summer, I'm also interning at Roche Pharmaceuticals, building graphical neural networks for cancer immunotherapy outcome & mechanism prediction.

In addition to fiddling with code and making art, I love petting my cat, playing social deduction games, spinning poi, and skateboarding!


Paint By Word

David Bau, Alex Andonian, Audrey Cui, YeonHwan Park, Ali Jahanian, Aude Oliva, Antonio Torralba
We investigated the problem of zero-shot semantic image painting — by using an image-text similarity network to guide the gradients of an image generation model, we "paint" selected regions to fit textual descriptions.
Paper submitted to a 2021 computer vision conference.

Built on Pytorch.

From Pixel to Paragraph: A Neural Artwork Analysis Paragraph Generator (Jun 2018 - Feb 2020)

Inspired by the subjectivity of art in my local art museum, I developed a model that generates paragraphs analyzing input pieces of artwork. I modified LeakGAN (an unconditional GAN) to condition on input artwork features so that it can learn image-text relationships. To make this more user friendly, I used Unity to develop an augmented reality interface (currently just a proof of concept).

Built with tensorflow, nltk, keras, numpy, BeautifulSoup, Unity, Microsoft Azure


Mural (2020)
sketchbook painting (2020)
Catlendar (2019)
Shower Triptych (2019)
Hyphenated American (2018)
Beyond the Horizon (2018)
Hearts in San Francisco (2019)
The Explorer and Her Cat (2018)
Nature with a Capital N (2018)
Chambers (2019)
Thumbs Up (2019)
Her Heart Beats to the Crunch of Shrimp Chips (2020)