Direct sampling is an efficient information theory-based multipoint simulation method. To reduce the computational cost, many approaches have been proposed to speed up this simulation method. In this paper, an edge-based two-stage strategy is proposed to achieve speed-up. The proposed method first performs a simulation on a coarse grid and then detects edge cells in the simulation grid. Next, only edge cells are simulated, and the remaining cells are assigned the average value of the neighbouring simulated and hard data cells. The proposed method needs only one easily-interpreted parameter and can be combined with other existing acceleration methods for direct sampling, such as parallelization. Comparison experiments are performed on categorical, continuous and multivariate variables, including two-dimensional and three dimensional examples. The experimental results show that the proposed method can reduce the simulation time by almost half with minimal loss in terms of pattern reproduction.

直接采样是一种有效的基于信息论的多点仿真方法。为了降低计算成本，已经提出了许多方法来加速该仿真方法。本文提出了一种基于边缘的两阶段策略来实现加速。所提出的方法首先在粗网格上执行仿真，然后在仿真网格中检测边缘单元。接下来，仅模拟边缘单元，其余单元被分配相邻模拟单元和硬数据单元的平均值。所提出的方法仅需要一个易于解释的参数，并且可以与其他现有的加速方法相结合以进行直接采样，例如并行化。对分类变量，连续变量和多元变量（包括二维和三维示例）进行比较实验。