AdaBits: Neural Network Quantization with Adaptive Bit-Widths
This work addresses the need for flexible and efficient neural network deployment on diverse hardware platforms, offering an incremental improvement in quantization methods.
The paper tackles the problem of deploying neural networks on platforms with varying resource constraints by enabling adaptive bit-widths for weights and activations, achieving comparable performance to individual models through joint training and a Switchable Clipping Level technique, with demonstrations on models like MobileNet-V1/V2 and ResNet-50 showing improved accuracy-efficiency trade-offs.
Deep neural networks with adaptive configurations have gained increasing attention due to the instant and flexible deployment of these models on platforms with different resource budgets. In this paper, we investigate a novel option to achieve this goal by enabling adaptive bit-widths of weights and activations in the model. We first examine the benefits and challenges of training quantized model with adaptive bit-widths, and then experiment with several approaches including direct adaptation, progressive training and joint training. We discover that joint training is able to produce comparable performance on the adaptive model as individual models. We further propose a new technique named Switchable Clipping Level (S-CL) to further improve quantized models at the lowest bit-width. With our proposed techniques applied on a bunch of models including MobileNet-V1/V2 and ResNet-50, we demonstrate that bit-width of weights and activations is a new option for adaptively executable deep neural networks, offering a distinct opportunity for improved accuracy-efficiency trade-off as well as instant adaptation according to the platform constraints in real-world applications.