Phasor-Driven Acceleration for FFT-based CNNs

TL;DR

This paper introduces a phasor-based approach to accelerate FFT-based CNNs, achieving significant speedups during training and inference without altering existing models.

Contribution

It proposes using the phasor form for FFT in CNNs, offering a more efficient spectral domain computation method that enhances speed.

Findings

Speed improvements up to 1.376x on CIFAR-10 during training

Speed improvements up to 1.390x on CIFAR-10 during inference

Applicable to existing CNN models without modifications

Abstract

Recent research in deep learning (DL) has investigated the use of the Fast Fourier Transform (FFT) to accelerate the computations involved in Convolutional Neural Networks (CNNs) by replacing spatial convolution with element-wise multiplications on the spectral domain. These approaches mainly rely on the FFT to reduce the number of operations, which can be further decreased by adopting the Real-Valued FFT. In this paper, we propose using the phasor form, a polar representation of complex numbers, as a more efficient alternative to the traditional approach. The experimental results, evaluated on the CIFAR-10, demonstrate that our method achieves superior speed improvements of up to a factor of 1.376 (average of 1.316) during training and up to 1.390 (average of 1.321) during inference when compared to the traditional rectangular form employed in modern CNN architectures. Similarly, when evaluated on the CIFAR-100, our method achieves superior speed improvements of up to a factor of 1.375 (average of 1.299) during training and up to 1.387 (average of 1.300) during inference. Most importantly, given the modular aspect of our approach, the proposed method can be applied to any existing convolution-based DL model without design changes.