ReBNet: Residual Binarized Neural Network

TL;DR

ReBNet introduces a reconfigurable binary neural network framework that enhances accuracy through residual binarization, enabling efficient FPGA deployment with adjustable throughput and minimal hardware overhead.

Contribution

It presents the first reconfigurable scheme for binary neural networks that improves accuracy without increasing hardware complexity.

Findings

Improves classification accuracy with residual binarization.

Maintains low hardware overhead despite multi-level features.

Provides a tradeoff between throughput and accuracy.

Abstract

This paper proposes ReBNet, an end-to-end framework for training reconfigurable binary neural networks on software and developing efficient accelerators for execution on FPGA. Binary neural networks offer an intriguing opportunity for deploying large-scale deep learning models on resource-constrained devices. Binarization reduces the memory footprint and replaces the power-hungry matrix-multiplication with light-weight XnorPopcount operations. However, binary networks suffer from a degraded accuracy compared to their fixed-point counterparts. We show that the state-of-the-art methods for optimizing binary networks accuracy, significantly increase the implementation cost and complexity. To compensate for the degraded accuracy while adhering to the simplicity of binary networks, we devise the first reconfigurable scheme that can adjust the classification accuracy based on the application. Our proposition improves the classification accuracy by representing features with multiple levels of residual binarization. Unlike previous methods, our approach does not exacerbate the area cost of the hardware accelerator. Instead, it provides a tradeoff between throughput and accuracy while the area overhead of multi-level binarization is negligible.