Create a timeline

Public timelines

jan 5, 2017 - Deep Learning with Low Precision by Half-wave Gaussian Quantization (HWGQ)

Description:

本文主要讨论了在低位宽量化的近似梯度训练中，梯度消失的问题，通过设计新的量化函数(sign/Q)和激活函数来保证梯度不消失，以实现量化模型的收敛。有趣的是，仅通过对relu函数的剪切或者log-tail变换，模型的性能得到了显著提升。这或许就是特征的影响

Added to timeline:

Quantization timeline

29 Mar 2020

1

0

744

Date:

jan 5, 2017

Now

~ 8 years and 5 months ago

About & Feedback Terms Privacy Library FAQ

Cabinet Get premium

Donate

The service accepts bank transfer (ACH, Wire) or cards (Visa, MasterCard, etc). Processed by Stripe.

Secured with SSL