The recent work on lattice-based space-filling designs
地点：Room 1303, Sciences Building No. 1
In this talk, He Xu will present several recent works of mine on constructing experimental design for computer experiments. Computer experiments are computer programs that simulate real systems using mathematical models. Some such simulations are extremely slow in computation, calling for experimental designs that are efficient in building surrogate models. Some recent work suggests that space-filling designs with low fill distance, high separation distance, and certain projective properties are appealing for computer experiments. Usually, such designs are generated using global optimization techniques. However, because the optimization problem is fundamentally difficult, near optimal designs are rarely obtained. His approach is to use lattices that have distance properties in Euclidean space to construct space-filling designs. Such designs include rotated sphere packing designs and interleaved lattice-based designs.
About the Speaker:
何煦，本科毕业于北京大学数学学院，博士毕业于威斯康星大学麦迪逊分校统计系。2012年起进入中国科学院数学与系统科学研究院，任助理研究员、副研究员。何煦的研究方向包括实验设计、计算机实验及不确定性量化，在Annals of Statistics，Journal of the American Statistical Association，Biometrika等杂志发表论文十余篇。