Patent · US Active

Fixed-point training method for deep neural networks based on static fixed-point conversion scheme

US11308392B2 · kind B2 · utility

1Cited by

2References

17Claims

0Family size

Assignee

XILINX TECHNOLOGY BEIJING LIMITED · CN

Inventors

Xin Li · Foshan, CN
Tong Meng · Beijing, CN
Song Han · Beijing, CN

Key dates

Filing date	Sep 1, 2017
Grant date	Apr 19, 2022
Priority date	—
Expiry date	Mar 1, 2040

Classification

Technology area (CPC G)Physics
CPC primaryG06N7/01
WIPO fieldComputer technology
WIPO sectorElectrical engineering

Abstract

The present disclosure proposes a fixed-point training method and apparatus based on static fixed-point conversion scheme. More specifically, the present disclosure proposes a fixed-point training method for LSTM neural network. According to this method, during the fine-tuning process of the neural network, it uses fixed-point numbers to conduct forward calculation. Accordingly, within several training cycles, the network accuracy may returned to the desired accuracy level under floating point calculation.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.