Efficient Winograd Convolution via Integer Arithmetic

Abstract

Convolution is the core operation for many deep neural networks. The Winograd convolution algorithms have been shown to accelerate the widely-used small convolution sizes. Quantized neural networks can effectively reduce model sizes and improve inference speed, which leads to a wide variety of kernels and hardware accelerators that work with integer data. The state-of-the-art Winograd algorithms pose challenges for efficient implementation and execution by the integer kernels and accelerators. We introduce a new class of Winograd algorithms by extending the construction to the field of complex and propose optimizations that reduce the number of general multiplications. The new algorithm achieves an arithmetic complexity reduction of $3.13$x over the direct method and an efficiency gain up to $17.37\%$ over the rational algorithms. Furthermore, we design and implement an integer-based filter scaling scheme to effectively reduce the filter bit width by $30.77\%$ without any significant accuracy loss.

9 pages, 5 figures, and quite a few matrices

References in corpus (3)

• TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
• Quantizing deep convolutional networks for efficient inference: A whitepaper
• Enabling Sparse Winograd Convolution by Native Pruning

Cited by in corpus (5)

• Searching for Winograd-aware Quantized Networks
• Fast Convolution based on Winograd Minimum Filtering: Introduction and Development
• Quantaized Winograd/Toom-Cook Convolution for DNNs: Beyond Canonical Polynomials Base
• Communication Bounds for Convolutional Neural Networks
• Efficient Residue Number System Based Winograd Convolution