| The objective of this thesis is to evaluate algorithms for finding optimal or near-optimal partitions of Poisson distributed data with respect to an objective function derived using Bayesian statistics. Four basic types will be explored: local search, tabu search, simulated annealing, and dynamic programming. The background information relevant to the problem will be introduced, and each algorithm will be explained in detail. Results of performance comparisons for one and two-dimensional data will be presented. In dimension one, an algorithm that provably finds an optimal solution to the problem using the principles of dynamic programming will be the standard by which all other algorithms are compared. The analysis of algorithms for the two-dimensional problem will focus on improvements over local search results offered by tabu search and simulated annealing. |
