Welcome to Ke Sun's Homepage.

I am a Machine Learning researcher.

My name is Ke Sun (孙科). I am currently a postdoctoral researcher at University of Pennsylvania, working with Prof. Qi Long and Prof. Weijie Su.

I was a postdoctoral fellow working with Prof. Susan A. Murphy at Harvard University in 2025. I received my Ph.D. degree in Statistical Machine Learning from University of Alberta in 2024, advised by Prof. Linglong Kong, also affiliated at Alberta Machine Intelligence Institute (Amii). I was fortunate to visit Prof. Chengchun Shi at the London School of Economics and Political Science in 2023. I obtained my MPhil in Data Science at Peking University in 2020 advised by Prof. Zhouchen Lin and Prof. Zhanxing Zhu. I earned my Bachelor's degree in finantial statistics and risk management from SouthWestern University of Finance and Economics in 2017.

Contact: kesun6[AT] upenn.edu / ke.sun[AT]pennmedicine.upenn.edu

Links: Google Scholar / CV / Github / LinkedIn

Research Interests

My research interest is reinforcement learning, especially the algorithmic foundations and development grounded in mathematical and statistical principles. My key research goal is to build agents with effective perception, memory, reasoning, planning, exploration, and adaptation in complex and unknown environments to achieve decision intelligence, with broad applications in games, robotics, control systems, healthcare, economics, and language.

[1] Algorithmic and Theoretical Foundations in Pure RL (RL as Problem Formulation):

- Uncertainty & Exploration: Distributional Learning and Risk Control, Entropy Regularization, Predictive Uncertainty
- Robustness & Adaptation: Environmental Nonstationarity, Safe and Robust RL, Transfer/Multi-task/Meta/Continual RL
- Offline & Causality: Imitation Learning, Offline and Hybrid RL, Causality for RL, Adaptive Experiment Design

[2] Foundations in RL for General Intelligence (RL as Optimization Tool vs Problem Formulation):

- Language Intelligence: Alignment, Reasoning, Post-training
- Physical Intelligence: Representation, Model-based RL/World Model, Generative/Self-supervised RL, Multi-agent RL
- Agentic Intelligence: Partial Observability, Safety, Hierarchical RL

Selected Preprints / Publications [Full Publications]

[6] Ke Sun, Linglong Kong, Hongtu Zhu, Chengchun Shi. ARMA-Design: Optimal Treatment Allocation Strategies for A/B Testing in Partially Observable Experiments. (Minor Revision in Journal of American Statistical Association, 2026)

[5] Ke Sun*, Hongming Zhang*, Jun Jin, Chao Gao, Xi Chen, Wulong Liu, Linglong Kong. Principled Fast and Meta Knowledge Learners for Continual Reinforcement Learning. International Conference on Learning Representations (ICLR), 2026. [Code]

[4] Ke Sun, Yingnan Zhao, Enze Shi, Yafei Wang, Xiaodong Yan, Bei Jiang, Linglong Kong. Intrinsic Benefits of Being Categorical Distributional: Uncertainty-aware Regularized Exploration in Reinforcement Learning. Advances in Neural Information Processing Systems (NeurIPS), 2025

[3] Ke Sun, Yingnan Zhao, Wulong Liu, Bei Jiang, Linglong Kong. Distributional Reinforcement Learning with Regularized Wasserstein Loss [code]. Advances in Neural Information Processing Systems (NeurIPS), 2024

[2] Ke Sun*, Yafei Wang*, Yi Liu, Yingnan Zhao, Bo Pan, Shangling Jui, Bei Jiang, Linglong Kong. (*equal contribution in alphabetical order). Damped Anderson Mixing for Deep Reinforcement Learning: Acceleration, Convergence, and Stabilization Advances in Neural Information Processing Systems (NeurIPS), 2021. [Code].

[1] Ke Sun, Zhanxing Zhu, Zhouchen Lin. AdaGCN: AdaBoosting Graph Convolutional Networks into Deep Models [code]. International Conference on Learning Representations (ICLR), 2021.

* denotes equal contribution

Favorite Quotes

Practice without theory is blind, but equally, theory without practice is mere intellectual play. ——Immanuel Kant

The statistical community has been committed to the almost exclusive use of data models. This commitment has led to irrelevant theory, questionable conclusions, and has kept statisticians from working on a large range of interesting current problems. …… The danger is that if we define the boundaries of our field in terms of familiar tools and familiar problems, we will fail to grasp the new opportunities. —— Leo Breiman

Talk is cheap, show me the code. —— Linus Torvalds