Causal Representation Learning for Controllable Text-To-Image Generation . This project aims to improve the controllability of existing text-to-image generation models. It expects to generate new know

Description

Causal Representation Learning for Controllable Text-To-Image Generation . This project aims to improve the controllability of existing text-to-image generation models. It expects to generate new knowledge for multimodal generative models by leveraging causal representation learning and contributes to the development of more trustworthy, interpretable, and controllable generative AI. Expected outcomes include the creation of a new large-scale multimodal benchmark dataset, the development of multiple practical controllable text-to-image generation models, and the establishment of theoretical guarantees for these models. This research will provide significant benefits by enabling safer AI applications across different domains and mitigating risks associated with unreliable or biased AI outputs.. Scheme: Discovery Early Career Researcher Award. Field: 4603 - Computer Vision and Multimedia Computation. Lead: Dr Yu Yao