Vertical Layering of Quantized Neural Networks for Heterogeneous Inference

TL;DR

This paper introduces a vertical-layered neural network weight representation enabling a single model to support multiple quantization precisions, reducing training and maintenance costs for heterogeneous device deployment.

Contribution

It proposes a novel vertical-layered weight representation and a once QAT scheme that encapsulate multiple quantized models into one, allowing on-demand precision adjustment.

Findings

Achieves comparable performance to dedicated quantized models.

Enables one-time training for multiple quantization levels.

Reduces costs in model training and maintenance for heterogeneous devices.

Abstract

Although considerable progress has been obtained in neural network quantization for efficient inference, existing methods are not scalable to heterogeneous devices as one dedicated model needs to be trained, transmitted, and stored for one specific hardware setting, incurring considerable costs in model training and maintenance. In this paper, we study a new vertical-layered representation of neural network weights for encapsulating all quantized models into a single one. With this representation, we can theoretically achieve any precision network for on-demand service while only needing to train and maintain one model. To this end, we propose a simple once quantization-aware training (QAT) scheme for obtaining high-performance vertical-layered models. Our design incorporates a cascade downsampling mechanism which allows us to obtain multiple quantized networks from one full precision source model by progressively mapping the higher precision weights to their adjacent lower precision counterparts. Then, with networks of different bit-widths from one source model, multi-objective optimization is employed to train the shared source model weights such that they can be updated simultaneously, considering the performance of all networks. By doing this, the shared weights will be optimized to balance the performance of different quantized models, thus making the weights transferable among different bit widths. Experiments show that the proposed vertical-layered representation and developed once QAT scheme are effective in embodying multiple quantized networks into a single one and allow one-time training, and it delivers comparable performance as that of quantized models tailored to any specific bit-width. Code will be available.