Deep Learningを用いた階層的な部分タスクグラフ検出法の構築

Abstract

Since task scheduling problems belong to a class of the strong NP-hard combinatorial optimization problem, the required scheduling time increases exponentially as the number of tasks increases. Therefore, we find some small subtask graphs that can be optimally scheduled in the overall task graph, and solve them individually as a scheduling problem. Then, each subtask graph can be treated as one macro task in the whole task graph. This reduces the apparent number of tasks in the overall task graph, reduces the scale of the task graph, and significantly reduces the search time that depends on the number of tasks. We call this hierarchical scheduling. However, this partial task graph detection has a drawback that it becomes a combinatorial optimization problem by itself. Therefore, in this paper, we construct and evaluate a method for detecting partial task graphs using deep learning.