|Title||Parameterized hierarchical annealing for scientific models|
|Publication Type||Conference Paper|
|Year of Publication||2004|
|Authors||Fieguth, P., and S. K. Alexander|
|Conference Name||2004 International Conference on Image Analysis and Recognition|
The accurate synthesis of binary porous media is a difficult problem. Initial applications of simulated annealing in this context with small data sets and simple energy functions have met with limited success. Simulated annealing has been applied to a wide variety of problems in image processing. Particularly in scientific applications such as discussed here, the computational complexity of this approach may constrain its effectiveness; complex, non-local models on large 2D and 3D domains may be desired, but do not lend themselves to traditional simulated annealing due to computational cost. These considerations naturally lead to a wish for hierarchical/multiscale methods. However, existing methods are few and limited. In this paper a method of hierarchical simulated annealing is discussed, and a simple parameterization proposed to address the problem of moving through the hierarchy. This approach shows significant gains in convergence and computational complexity when compared to the simulated annealing algorithm.