Yael Vinker
Tel Aviv University
yaelvinker@mail.tau.ac.il
Bio
Yael Vinker is a PhD student at Tel Aviv University, advised by Prof. Daniel Cohen-Or and Prof. Ariel Shamir, and will soon join MIT as a postdoctoral researcher under the supervision of Prof. Antonio Torralba. Her research lies at the intersection of computer graphics and art, with a focus on generative models. Yael has been recognized with two SIGGRAPH Best Paper Awards and an Honorable Mention Award for her works CLIPasso, Inspiration-Tree, and Word-as-Image. Her additional honors include being nominated for the Fulbright Postdoctoral Fellowship, receiving the Blavatnik Prize for outstanding Israeli doctoral students in computer science, and being awarded the VATAT Scholarship for outstanding female PhD candidates. She earned both her BSc and MSc degrees from The Hebrew University of Jerusalem.
Areas of Research
- Computer Graphics and Vision
Generative Visual Communication in the Era of Vision-Language Models
Visual communication, dating back to prehistoric cave paintings, is the use of visual elements to convey ideas and information. In our visually saturated world, effective design demands an understanding of graphic design principles, visual storytelling, human psychology, and the skill to distill complex information into clear visuals. This dissertation explores how computational models can be leveraged to automatically produce effective visual communication designs, especially in light of recent advancements in large vision-language models (VLMs) and generative ‘AI’. The primary objectives are: (1) leveraging the hidden representations of VLMs to tackle cognitive-visual tasks from a design perspective, and (2) developing practical generative tools to support designers. Despite impressive capabilities in generating high-quality images from textual descriptions, existing generative models face challenges. They struggle with distilling complex ideas into abstract visuals and are limited to pixel-based representations, which are suboptimal for many design tasks. In this dissertation, we explore various aspects of visual communication, including sketches, visual abstraction, typography, animation, and visual inspiration. We develop computational approaches to address new generative tasks within these domains. The key idea is to leverage the strong priors of pretrained VLMs using inference-time optimization, imposing constraints on their operational space, and introducing task-specific regularizations.