Ho Fung Leung

Also known as Ho-fung Leung

Ho Fung Leung is an Emeritus Professor of computer science and engineering at The Chinese University of Hong Kong, whose research on multiagent systems covers reinforcement learning, game-theoretic analysis and the learning dynamics of agent populations. He received BSc and MPhil degrees from CUHK and a PhD from the University of London with the Diploma of Imperial College, chaired the Hong Kong Chapter of ACM in 1998, and served as Chairman of the CUHK Department of Computer Science and Engineering from 2011 to 2014. His papers include a 2013 IJCAI study with Jianye Hao on social learning among cooperative agents, a 2019 NeurIPS paper that used mean field theory to model multiagent Q-learning, and a 2023 NeurIPS analysis of offline reinforcement learning with options. He has authored more than 300 publications, including 5 research monographs and 5 edited volumes.

Now at: Chinese University of Hong Kong
Born: 1963

Years link to their sources.

Timeline

1998

Chairing ACM Hong Kong

He holds BSc and MPhil degrees in computer science from The Chinese University of Hong Kong and a PhD from the University of London, with the Diploma of Imperial College in computing. In 1998, he chairs the Hong Kong Chapter of ACM and publishes "Extending GENET to Solve Fuzzy Constraint Satisfaction Problems" at AAAI with Jason H. Y. Wong. The paper extends GENET to fuzzy constraint satisfaction problems.
2009

Ontology learning from folksonomies

In 2009, he coauthors "Towards Ontology Learning from Folksonomies" at IJCAI with Jie Tang, Qiong Luo, Dewei Chen and Jibin Gong. The paper proposes a learning approach that builds an ontology capturing the hierarchical semantic structure of folksonomies.
2011–2014

Department chairman at CUHK

He is Chairman of the Department of Computer Science and Engineering at The Chinese University of Hong Kong from 2011 to 2014.
2013

Social learning in cooperative agents

In 2013, he and Jianye Hao publish "The Dynamics of Reinforcement Social Learning in Cooperative Multiagent Systems" at IJCAI. The paper studies a large population in which each agent meets a randomly chosen partner in every round, and compares individual action learners with joint action learners.
2019

Mean field Q-learning dynamics

In 2019, he coauthors a NeurIPS paper with Shuyue Hu and Chin-wing Leung on multiagent Q-learning in repeated symmetric games. Using mean field theory, they derive a Fokker-Planck equation for the distribution of Q-values in a population of agents, which describes the Q-learning dynamics of n agents with a system of only three equations.
2022

Q-learning with local interaction

In 2022, he coauthors an IJCAI paper with Chin-wing Leung and Shuyue Hu on Q-learning under the generalized social learning protocol. They model each agent's Q-values as a system of stochastic differential equations and express the population's Q-value distribution with a Fokker-Planck equation.
2023–2025

Offline RL and PAK-UCB

In 2023, he and Xiaoyan Hu give the first analysis of the sample complexity of offline reinforcement learning with options at NeurIPS, and propose the PEVIO algorithm. In 2025, he coauthors the ICML paper "PAK-UCB Contextual Bandit" with Xiaoyan Hu and Farzan Farnia, an online learning framework that predicts the best generative model or LLM for a given text prompt.
As of October 2026

Emeritus Professor at CUHK

He is an Emeritus Professor in the Department of Computer Science and Engineering at The Chinese University of Hong Kong. His research interests include multiagent systems (reinforcement learning, emergence phenomena and evolution dynamics), game theoretic analysis, ontologies and big data analytics, and he has authored more than 300 publications, including 5 research monographs and 5 edited volumes.