Dual Precision Deep Neural Network

TL;DR

This paper introduces a dual-precision deep neural network that enables on-line switching between precision modes without re-training, balancing accuracy and complexity during inference.

Contribution

It proposes a novel dual-precision DNN architecture with a two-phase training process for simultaneous optimization of both precision modes.

Findings

Supports on-line precision switching without re-training

Optimizes both low- and high-precision modes effectively

Enhances inference flexibility and efficiency

Abstract

On-line Precision scalability of the deep neural networks(DNNs) is a critical feature to support accuracy and complexity trade-off during the DNN inference. In this paper, we propose dual-precision DNN that includes two different precision modes in a single model, thereby supporting an on-line precision switch without re-training. The proposed two-phase training process optimizes both low- and high-precision modes.