HBM: A Suitable Neural Network Pruning Technique to Optimize the Execution Time of the Novel Neural Network Controller (NNC) that Eliminates Buffer Overflow

摘要

In this paper the HBM optimization technique is proposed for pruning the extant NNC (Neural network Controller) for shorter execution time. Although the NNC eliminates buffer overflow effectively, it is not suitable for time-critical applications because of its long execution time, which is the time required to converge to the target optimal or reference point of the objective junction. The HBM differentiates the important neural network connections from the unimportant one in a relative manner. Afterwards, those parameters that are found to have less impact on the convergence process toward the optimal point would be eliminated. The experimental results with the optimized NNC show that the HBM technique is indeed effective, as demonstrated by the timing analysis exercise with the VTune Performance Analyzer.