An Exact Dynamic Programming Based Method to Solve Optimisation Problems Using GPUs

In this paper we present a dynamic programming based technique that is suitable for providing exact solutions to a subset of optimisation problems, using general purpose GPU computing. The primary feature of this model is to efficiently use the computational and memory resources of the GPU, whilst still remaining abstract enough to allow implementation on a variety of problems. Secondly, as exact dynamic programming methods are often limited by memory complexity, great consideration has been given to reducing this constraint, allowing large scale problems to be solved. To demonstrate the effectiveness of the proposed model we test it against three problems, the 0-1 knapsack problem, the longest common subsequence problem, and the travelling salesman problem.