2020年2月29日 · We propose a novel sinusoidal regularization, called SINAREQ, for deep quantized training. Leveraging the sinusoidal properties, we seek to learn multiple quantization parameterization in conjunction during gradient-based training process.

(ICLR2021) WaveQ: Gradient-Based Deep Quantization of Neural Networks through Sinusoidal Adaptive Regularization 深度量化可能会导致重大的准确性损失。 由于层之间的相互依赖性很强，并且在同一网络上表现出不同的特征，因此选择最佳的每层粒度位宽并不是一件简单的事情。

We propose a novel sinusoidal regularization, called WaveQ, for deep quantized training. Leveraging the sinusoidal properties, we seek to learn multiple quantization parameterization in conjunction dur-ing gradient-based training process.

WaveQ is a sinusoidal-based quantization-aware regularization method. Adding our parametrized sinusoidal regularizer enables us to not only find the quantized weights but also learn the bitwidth of the layers by making the period of the sinusoidal regularizer a trainable parameter.

We show that WaveQ balances compute efficiency and accuracy, and provides a heterogeneous bitwidth assignment for quantization of a large variety of deep networks (AlexNet, CIFAR-10, MobileNet, ResNet-18, ResNet-20, SVHN, and VGG-11) that virtually pre-serves the accuracy. WaveQ is versatile and can also be used with predetermined

2021年1月1日 · We show how WaveQ balance compute efficiency and accuracy, and provide a heterogeneous bitwidth assignment for quantization of a large variety of deep networks (AlexNet, CIFAR-10, MobileNet, ResNet-18, ResNet-20, SVHN, and …

We show how SINAREQ balance compute efficiency and accuracy, and provide a heterogeneous bitwidth assignment for quantization of a large variety of deep networks (AlexNet, CIFAR-10, MobileNet, ResNet-18, ResNet-20, SVHN, and VGG-11) that virtually preserves the accuracy.

As deep neural networks make their ways into different domains and application, their compute efficiency is becoming a first-order constraint. Deep quantization, which reduces the bitwidth of the operations (below eigh…

This paper proposes WaveQ, a content-based image retrieval system that classifies images as texture or non-texture, then uses a Daubechies wavelet decomposition to extract feature vector information from the images, and finally applies the OPTICS clustering algorithm to cluster the extracted data into groups of similar images.

WaveQ: Gradient-Based Deep Quantization of Neural Networks through Sinusoidal Adaptive Regularization arXiv - CS - Machine Learning Pub Date : 2020-02-29, DOI: arxiv-2003.00146