报告题目:Constrained Boolean Pareto Optimization: Theories and Practical Algorithms
报告人:钱超博士,中国科学技术大学副研究员
主持人:周爱民
报告时间 :2016年9月29日10:00-11:00
报告地点:中山北路3663号华东师范大学理科大楼B816室
报告摘要:
Optimization tasks usually come with constraints, which must be satisfied by the final solutions. Particularly, for constrained Boolean optimization, i.e., the solution space is , the problem is often NP-hard. Although there are some approximation algorithms, e.g., greedy methods, convex relaxation methods and heuristic search methods, it has been found that their performances are limited in many cases. In this talk, we will introduce a new approach for constrained Boolean optimization, called Pareto optimization. This method first reformulates the original problem as a bi-objective optimization problem, which requires optimizing the original objective function and minimizing the constraint violation degree simultaneously. The bi-objective optimization problem is then handled by a randomized iterative algorithm. From the solved Pareto set of solutions, the best solution satisfying the constraints is selected as the final solution to the original constrained problem. We will introduce both theoretical and practical results on Pareto optimization, and further introduce a parallel version of Pareto optimization, which can achieve linear speedup while preserving the approximation quality.
报告人简介:
钱超,博士,中国科学技术大学副研究员。主要研究领域为人工智能、机器学习和演化计算,特别关注演化计算的理论基础与多目标演化学习。分别于2009年6月和2015年9月获得南京大学计算机系学士学位和博士学位,2015年11月加入中国科学技术大学计算机学院中国科大-伯明翰大学智能计算联合研究所从事科研工作。在Artificial Intelligence、Evolutionary Computation、IEEE Transactions on Evolutionary Computation、NIPS、IJCAI、AAAI等国际一流期刊和顶级会议上发表多篇论文,其中在演化计算重要国际会议ACM GECCO 2011上的论文获最佳理论论文奖。曾获PAKDD 2012数据挖掘竞赛冠军,2014年IBM全球博士生英才计划奖,2016年CCF-腾讯犀牛鸟科研基金。担任了IJCAI’15、GECCO’16、AAAI’17等国际会议的程序委员,是国际期刊IEEE TEvC、IEEE TNNLS、IEEE CIM等的审稿人。