I am a third-year undergraduate student at Nanyang
Technological University (NTU), Singapore. My research interests lie in
Generative AI, with applications in world models and robotics.
I am currently doing research on 3D/world models at the Spatial Intelligence Lab, advised
by Prof. Jiawang Bian.
I am also an algorithm engineer intern at ByteDance, working on AIGC, and a Research Consultant at
WorldQuant.
AllocFlow reads the raw price history of every asset and generates the whole long/short portfolio
at once with a cross-asset flow Transformer, trained by imitation and then group-relative RL. After
fees it reaches a Calmar ratio of 4.04 on 14 unseen months of crypto futures, and 4.90 on six more
months without retraining.
USS lets a robot follow a target specified by text, a point, a box or a mask, mapping any of
these prompts to egocentric waypoints end to end. Policies trained only in simulation transfer
zero-shot to a real humanoid robot across 320 real-world trials.
DN-MOPD keeps label routing in multi-teacher on-policy distillation but rescales each domain's
teacher feedback by its measured spread, so that no single domain dominates the shared student.