Statistics Seminars: Space filling designs in unusual subspaces
3 February 2012 15:00 in CM221
The analysis of computer experiments relies on experimental designs that fill space with a minimum number of points. Space filling designs in rectangular regions 9in many dimensions) are fairly well understood, varieties of Latin Hypercubes and quasi-Monte Carlo sequences. However producing a design that fills space in a non- rectangular area is much more difficult. The introduction of ideas such as history matching and implausibility have raised an even more difficult problem can we make space filling designs that fill spaces that cannot be specified in advance and have no 'nice' geometric properties a priori. After giving a brief survey of space filling designs I will outline a new design that should allow us to produce space filling designs in any sub-space defined only by a membership function.
Contact email@example.com for more information