My work is, in general, about fundamental research on computational approaches for Learning and Optimization, two most important problems in Artificial Intelligence. I'm also frequently attracted by other relevant domains, such as Smart Logistics, Structural and Multi-Disciplinary Optimization and Computational Finance, for which applied research is required to produce application-oriented learning and optimization techniques.

Most of my research could also be viewed as arising from Evolutionary Computation, which is essentially a distributed heuristic search framework widely applicable for modelling, learning and optimization problems, especially for hard problems where limited prior knowledge is available.

Selected topics for fundamental research are listed below.

1. Scalable Evolutionary Search

Research in this direction aims at systematically boosting the capacity of Evolutionary Computation on problems with huge search space, which has been believed as a major challenge for most EAs. Approaches for this purpose include:

• Co-evolutionary Search: Introducing the divide-and-conquer idea to guide EAs adaptively search different regions of the search space

• Parallel Algorithm Portfolios: Leveraging on high performance computing to enhance both the extreme performance and reliability of EAs, without suffering the wall-clock runtime but only computational resources.

• Surrogate-assisted Search: Exploiting data generated during the search course to alleviate the cost of evaluating a future solution.

2. Reinforcement and Evolutionary Learning

Reinforcement Learning is a learning problem that lies exactly in the "backyard" of EAs, because the objective function of most RL tasks so far rely on a noisy and non-differentiable simulator. Thus it’d be quite interesting to see whether EC could offer a promising alternative approach for RL.

3. Learning and Optimization with Uncertainty

Uncertainty is ubiquitous in real-world learning and optimization tasks. It could be due to the dynamically changing physical world, the noise caused by imprecise measurements, or even the unpredictable nature of human behaviors. We are specifically interested in new learning/optimization methods that could handle various forms of uncertainty. This has led to exploration on the following topics:

• Incremental learning with concept drift

• Evolutionary computation for Dynamic optimization

• Learning from crowds (Crowdsourcing Learning)

计算机导论A、人工智能、研究方法、智能数据分析、创新教学实践

从演化计算、机器学习的基础理论出发，完善了数据驱动的智能算法设计手段，围绕复杂学习与优化问题，形成了可扩放演化搜索方法体系，并针对智慧物流、无人系统、边缘智能等领域的典型问题提出了一系列高效的算法。在IEEE/ACM Trans、CCF-A类学术会议发表论文60余篇。更多信息请点击此处。