Yicong Hong

I am a researcher at a Stealth Startup.

Recently, my work has focused on building interactive video world models, pursuing extended duration (Progressive Diffusion), long-horizon memory (Test-Time Training), consistent geometry (WorldCam), high modeling efficiency (Hydra-Transformer-Hybrid), and real-time control (RELIC).

See our latest work — The RELIC World Model !!!

I believe this direction paves the way for creating immersive virtual worlds and supports embodied intelligence in learning and reasoning about the real world.

Prior to this, I created the Recurrent Transformers and the Large Reconstruction Models — two projects I'm particularly proud of.

I completed my Ph.D. (Dec 2023) at the CECC at the Australian National University, advised by Prof. Stephen Gould, Prof. Qi Wu, and Prof. Lexing Xie.