BCNN: A Binary CNN with All Matrix Ops Quantized to 1 Bit Precision

TL;DR

This paper introduces a fully binary convolutional neural network (BCNN) that uses 1-bit matrix operations, achieving competitive accuracy on ImageNet with optimized training and implementation strategies.

Contribution

It presents a novel binary CNN architecture with all matrix operations quantized to 1-bit, demonstrating universality and practical training methods.

Findings

71.24% top-1 accuracy on ImageNet

Two-step training procedure developed

Optimized implementation strategies for binary operands

Abstract

This paper describes a CNN where all CNN style 2D convolution operations that lower to matrix matrix multiplication are fully binary. The network is derived from a common building block structure that is consistent with a constructive proof outline showing that binary neural networks are universal function approximators. 71.24% top 1 accuracy on the 2012 ImageNet validation set was achieved with a 2 step training procedure and implementation strategies optimized for binary operands are provided.