Abstract:
Deep generative models make visual content creation more accessible to novice and professional users alike by automating the synthesis of diverse, realistic content based on a collected dataset. People often use generative models as data-driven sources, making it challenging to personalize a model easily. Currently, personalizing a model requires careful data curation, which is too time-consuming and costly for everyday users. Hence, my research directs towards enabling billions of everyday users to easily collaborate, create, and share their personalized models without ML expertise.

In my talk, I present two directions towards achieving this goal: (1) “human-in-the-loop model creation”: bypassing the bottleneck of data collection process, and enable users to directly create models using simple interfaces (e.g., sketches, control points), (2) “content-based model search”: building a search engine “Modelverse”, where neural-network-based models are indexed and searchable, allowing users to search and share personalized models, creating a whole new community of content creators.

Committee:
Jun-Yan Zhu
Shubham Tulsiani
Deepak Pathak
Yufei Ye