Jet New

📄 Resume

💻 I’m a final year computer science undergraduate at the National University of Singapore, enrolled in the Turing Research Programme and University Scholars Programme.

♟️ I’m currently researching object-centric world models for multi-agent reinforcement learning, advised by Asst Prof Harold Soh and Prof Lee Wee Sun.

💼 I previously interned as a data scientist at Indeed and machine learning engineer at Grab.

🧠 I run an AI interest group with ~200 people, to meet new people with similar interests in AI.

news

Feb 1, 2022	I will be interning at Indeed as Data Scientist in the coming Summer 2022.
Jan 1, 2022	I’ll be researching graph neural networks for model-based multi-agent reinforcement learning under Prof Lee Wee Sun and Asst Prof Harold Soh.
Dec 1, 2021	My CS4246 project on graph neural networks for multi-agent reinforcement learning won the internal class project competition.
Aug 1, 2021	I was awarded Silver in the NUS School of Computing Student Awards for Accomplishment.

papers

Paper

Representation Learning for Model-Based Reinforcement Learning: A Survey

New, Jun Jie

2021

Abs PDF

Advancements in reinforcement learning and deep learning have surpassed human experts and achieved impressive feats on many complex sequential decision making problems. To tackle the high sample complexity of reinforcement learning methods, model-based reinforcement learning learns a model of the environment dynamics to improve the sample efficiency of learning. The learnt model also enables future prediction for forward planning to improve task performance. However, from the complex and high-dimensional nature of problems emerges a trade-off between the model’s sample efficiency and prediction accuracy. Representation learning methods are proposed to learn compressed yet informative latent representations of the state space. This survey aims to provide an overview of representation learning methods for learning and planning in model-based reinforcement learning, and discuss challenges and future directions from the present onward.