A Neural Network Based Algorithm Selector for Radar Task Scheduling
Zhen Qu, Zhen Ding, Peter W. Moo · 2020
A neural network based algorithm selector, to choose the most appropriate scheduling algorithm, is proposed in this paper. The approach uses the recurrent neural network (RNN) to learn and to select. The earliest start time algorithm and the random shifted start time algorithm, are considered in the RNN. The network is trained with 400,000 samples, and validated with 40,000 samples, resulting in a correct selection rate of 92%. The evaluation is done by numerical simulations, and the result shows an improved overall performance in terms of the schedule cost. The selection approach takes about 11 ms, thus it is practical for real world applications.