I obtained my M.Eng. and B.A. degrees in Engineering from the University of Cambridge under the supervision of Dr. Samuel Albanie. Previously, I completed my A-Levels at Harrow International School Hong Kong. My current research interest lies in text-to-image generation models, especially in introducing greater controllability into their generation processes.
I believe there is great area for improvement for text-to-image models to generate images that better fits human requirements, including: detailed human specification (indicating layout, object sizes and colours etc), incorporation of a customised subject into subsequent generations, edition of an existing image based on human instruction, and so on.
I am also concerned about copyright issues of text-to-image generation models, and interested in working on identification of cases when generated image overly resembles training images, model-unlearning to eliminate knowledge acquired from copyrighted training data without retraining the entire model, as well as techniques such as Anti-DreamBooth to protect images from unauthorised learning.