Weijie Zheng

Weijie2.jpg

Contact

Mail: zhengwj13@mails.tsinghua.edu.cn

Room: 3-126, FIT Building, Tsinghua University, Beijing, 100084

Current State

I am currently a PhD candidate in Computer Science and Technology Department in Tsinghua University.

Experience

  • 2013- now Tsinghua University, Beijing PhD, CS
  • 2009-2013 Harbin Institute of Technology, Harbin B.E., Mathematics

Research

I am interested in the optimization methods. I would like to figure out the effect of each element in the whole process, and be keen to make myself clear the essential start point of the design of existing methods, and eventually propose some new algorithms.

I have done some work on the Differential Evolution (DE), one kind of Evolutionary Algorithm, and provide a new perspective to DE variant designs that consider the randomness-reduced differential vector.

Publication

[1]Weijie Zheng, Haohuan Fu, and Guangwen Yang. "Targeted Mutation: A Novel Mutation Strategy for Differential Evolution." 27th International Conference on Tools with Artificial Intelligence (ICTAI). IEEE, 2015.[PDF][CITE]

Abstract: Different from the general DE variants that maintain the randomness of the differential vector, this paper attempts injecting the certainty and makes the search direction more effectively. The proposed basic mutation shows a competitive performance against two popular basic mutations on several benchmark functions, which leads a new perspective to design DE algorithms.

[2] Weijie Zheng, Haohuan Fu and Guangwen Yang. "TADE: Tight Adaptive Differential Evolution." 14th International Conference on Parallel Problem Solving from Nature (PPSN). Springer, 2016. [PDF][Supplyment][Matlab][CITE]

Abstract: This paper designs an adaptive DE which considers the randomness-reduced differential vector. Since reducing randomness in differential vector results in the large reduction on search area, its corre- sponding mutation is designed as a radical but effective pioneer to drive the minor subpopulation of the generation, and the majority employs a same base but foresight mutation as an officer to ensure the good exploration. Based on the success memory parameter adaption, an exchange between two subpopulations is designed to prolong the coming time of parameter homogeneity and save more time for exploration. Extensive experiments shows the competitive performance with five state-of-the-art DE variants.