Research Topics


My research interests evolve over time. Summarized below are some research questions that attracted me the most during the past few years.



Scalable Evolutionary Optimization

Scalability is a key issue for almost all computer algorithms. However, most meta-heuristic algorithms, e.g., evolutionary algorithms, scale poorly with the number of decision variables involved in an optimization problem. Hence, we have been working to develop novel scalable evolutionary algorithms, which could be used to tackle large-scale optimization problems. Specifically, (unlike some more classical computational problem, e.g., sorting) we are interested in problems that are not separable (decomposable) and that are characterized with non-continuous, non-differentiable, or even black-box objective functions. The core idea of our efforts along this direction is to explore smart method to decompose a problem in to sub-problems, such that a good solution to the original problem could be obtained by solving the related sub-problems. In short, our research can be regarded as a counter-part of divide-and-conquer method for non-separable large-scale (black-box) problems.

Selected relevant papers:



Learning and Optimization in Uncertain Environments

We are living in a world of uncertainty. This fact is, more often than not, exciting to me, since uncertainty may means some possibilities that had never come into my mind before. Hence, I've been always attracted by research topics that are relevant to the handling of uncertainties. Such topics (in various background) include, but not limited to:

Imbalanced Data Classification, where the "cost" of making mistakes on rare data is unknown or uncertain.
Evolutionary Robust/Dynamic Optimization, where the objective function and constraints may subject to uncertainty or even change over time.
Incremental Learning (highly relevant to online learning and stream data mining), where the concept (e.g., the underlying distribution of the data) may change over time.



Data Driven Meta-heuristic Search

Meta-heuristic search algorithms iteratively generate and tests candidate solutions to a problem, and hence can be viewed as a data generating process itself. The generated data consist quite useful information about the problem to solve and the interaction between algorithms and problems. Making the best use of such data is a methodology that has the potential to boost the performance of the state-of-the-art meta-heuristic search algorithms. We've carried out quite a lot research along this direction, particularly to develop novel evolutionary algorithms for complex problems (e.g., large-scale problems or problems subject to uncertainty) that cannot be satisfactorily tackled by conventional EAs.

Selected relevant papers:



Capacitated Arc Routing

Capacitated Arc Routing Problems (CARPs) is a hard combinatorial problem that has broad applications in logistic and transportation domains. We have developed a number of novel EAs to various versions of CARP. Most of them have achieved the best-known solutions on a large variety of benchmark problems.

Selected relevant papers:


[My Home]