Jing Yu LIMI'm a Doctoral candidate at NUS School of Computing , with affiliations at NUS Yong Loo Lin School of Medicine, NUS AI Institute and NUS Advanced Robotics Center. I research on World Models, Model-Based Reinforcement Learning, Representation Learning and Diffusion Models at the Medical Computing Lab and Cognitive AI for Science Lab supervised by Tze-Yun Leong and Dianbo Liu
Contact: jy_lim[at]comp.nus.edu.sg
|
|
ResearchI am interested in fundamental AI problems related to , world models, reinforcement learning and representation learning. |
|
JEDI: Latent End-to-end Diffusion Mitigates Agent-Human Performance Asymmetry in Model-Based Reinforcement Learning
Jing Yu LIM, Zarif Ikram, Samson Yu, Haozhe Ma, Tze-Yun Leong, Dianbo Liu Preprint., 2025 arXiv / We uncover an performance assymetry of existing Model-Based RL agents on the Atari100k benchmark across agent-optimal and human-optimal tasks; this is especially pronounced in pixel-based agents. To address this, we propose Joint Embedding DIffusion (JEDI) World Model which captures both the visual modeling power of diffusion models as well as the action temporal reasoning capabilities of latent world models. JEDI World Model agents performs more holistically across both sets of tasks. |
|
Latent Emission-Augmented Perspective-Taking (LEAPT) for Human-Robot Interaction
Kaiqi Chen, Jing Yu LIM, Kingsley Kuan, Harold Soh IROS, 2023 arXiv / We propose a deep multi-modal latent world model that enables the learning of uncertainty in latent space and perspective-taking of human’s observation and belief in partially observable environments. |
|
|