1、机器学习遇上运筹优化:一种双层优化方法A Bi-Level Framework for Learning to Solve Combinatorial Optimization on GraphsRunzhong Wang Zhigang Hua Gan Liu Jiayi Zhang Junchi Yan Feng Qi Shuang Yang Jun Zhou Xiaokang YangNeurIPS 2021Code Available at https:/ BioEducation 2019-now:PhD student at Shanghai Jiao Tong Universit
2、y 2015-2019:Bachelor student at Shanghai Jiao Tong UniversityResearch Interest Graph learning,especially machine learning and combinatorial optimization.Publications I have published 11 papers(7 first-author)in top conferences and journals,including TPAMI,NeurIPS,ICML,ICLR,ICCV,and CVPR.I develop an
3、d maintain several Github repositories with 1700+stars.Academic/Social Services I serve as the reviewer of ICML,NeurIPS,ICLR,CVPR,ICCV,AAAI,and MM.I serve as the class monitor of Wu Wen-Tsun Honorary Doctoral Class at Shanghai Jiao Tong University.At the oral session at ICCV 20193What is Optimizatio
4、n?optimization/ptmzen/nounthe action of making the best or most effective use of a situation or resource.(Oxford English dictionary)Convex OptimizationStrong general solversMany are polynomial-timeNon-Convex OptimizationNo general solversUsually exponential-time4What is Combinatorial Optimization?mi
5、n ,.discrete constraints on What Combinatorial Optimization(CO)looks like:for example:is binary has no more than nonzero elements is a path from A to B0/1 knapsackcardinality-constrained portfoliorouting5The Problems Considered in This Paper Graph Edit Distance:Minimize graph edit cost Hamiltonian C
6、ycle:Find valid Hamiltonian Cycles DAG Scheduling:Minimizemakespan timeMao et al.“Learning scheduling algorithms for data processing clusters.SIGCOMM 2019.Wang et al.“Combinatorial learning of graph edit distance via dynamic embedding.”CVPR 2021MethicillinPenicillin6Combinatorial Optimization Proble