Systems and methods for limiting data storage requirements of neural networks

Abstract

A split, sequential convolution method limits memory requirements for neural networks, such as convolutional neural networks (CNNs), by reducing temporal computational complexity and serializing operations. For example, a processor subdivides a plurality of channels in the input feature data for the CNN into groups of intermediate channels and processes each individual, subdivided group of intermediate channels sequentially rather than processing all of the channels simultaneously. Another method limits system memory accesses and improves overall efficiency by storing intermediate results in a buffer. For example, a processor limits data storage requirements of neural networks by storing intermediate outputs of convolutions configured to process input feature data in a local buffer rather than in system memory.

Dependencies are inferred from the claim text.