Minimal Filtering Algorithms for Convolutional Neural Networks

In this paper, we present several resource-efficient algorithmic solutions regarding the fully parallel hardware implementation of basic filtering operation performed in convolutional layers convolution neural networks. fact, these operations calculate two inner products neighboring vectors formed by a sliding time window from current data stream with an impulse response M-tap finite filter. We used extension Winograd’s minimal method and applied it to develop hardware-oriented algorithms for implementing M = 3, 5, 7, 9, 11. A proposed each case gives approximately 30% savings number embedded multipliers compared naive calculation methods.