My research focuses on the following key topics.

Game-theoretic Learning with Bandit Feedback

Problems in computer networks often arise from the interactions among multiple agents. In this research, we theoretically analyze scenarios involving multiple agents engaged in normal-form or extensive-form games with incomplete information and bandit feedback, employing uncoupled learning techniques. Specifically, we design bandit algorithms that guarantee convergence to equilibrium when adopted by all agents in the game, leveraging swap-regret-minimization techniques. Additionally, we explore the use of adaptive (optimistic or predictive) learning techniques to accelerate the convergence rate of these bandit algorithms.

Related Publications:

Zhiming Huang, Jianping Pan 

     pdf

     Zhiming Huang, Jianping Pan

     pdf

Zhiming Huang, Jianping Pan

     pdf poster video 


Combinatorial Learning with Bandit Feedback

Combinatorial optimization is also an important topic in computer networks. For example, a routing path selection involves a combination of feasible links. In this research, we are interested in the online optimization problem with bandit feedback, using Thompson sampling or Follow-the-regularized-learder (FTRL) techniques.

Related Publications:

Zhiming Huang, Jianping Pan

     pdf

     Zhiming Huang, Bingshan Hu, Jianping Pan

     pdf

     Zhiming Huang, Bingshan Hu, Jianping Pan

     pdf slides Award

Applications of Learning in Computer Networks

We apply the tools we studied above to concrete network problems, such as end-to-end congestion control, heterogenous network selection, caching and routing problems.

Related Publications:

     Zhiming Huang, Jianping Pan

     pdf

Zhiming Huang, Kaiyang Liu, Jianping Pan

pdf slides

     Zhiming Huang, Yifan Xu, Jianping Pan

     pdf

     Zhiming Huang, Bingshan Hu, Jianping Pan

     pdf