33
/de/
AIzaSyAYiBZKx7MnpbEhh9jyipgxe19OcubqV5w
August 1, 2025
3251714
288434
1
Public Timelines
Public TimelinesFAQ
Public TimelinesFAQ
For educational institutions For teachers For students Cabinet
For educational institutions For teachers For students Open cabinet
Erstellen
Create a timeline
Public timelines
Library
FAQ

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

Beschreibung:

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

Zugefügt zum Band der Zeit:

Quantization timeline

By龚成
29 Mär 2020
1
0
745

Datum:

5 Jan 2017 Jahr
Jetzt
~ 8 years and 5 months ago
About & FeedbackVereinbarungPrivatheitBibliothekFAQ
Support 24/7  
CabinetGet premium
Donate
The service accepts bank transfer (ACH, Wire) or cards (Visa, MasterCard, etc). Processed by Stripe.
Secured with SSL