![一种面向轻量级卷积神经网络的权值数值定点量化方法](/CN/2019/1/201/images/201911008463.jpg)
基本信息:
- 专利标题: 一种面向轻量级卷积神经网络的权值数值定点量化方法
- 专利标题(英):Weight value fixed-point quantification method for lightweight convolutional neural network
- 申请号:CN201911008463.1 申请日:2019-10-22
- 公开(公告)号:CN110837890A 公开(公告)日:2020-02-25
- 发明人: 杨晨 , 李博文 , 耿龙飞 , 王逸洲 , 耿莉
- 申请人: 西安交通大学
- 申请人地址: 陕西省西安市咸宁西路28号
- 专利权人: 西安交通大学
- 当前专利权人: 西安交通大学
- 当前专利权人地址: 陕西省西安市咸宁西路28号
- 代理机构: 西安通大专利代理有限责任公司
- 代理人: 郭瑶
- 主分类号: G06N3/08
- IPC分类号: G06N3/08 ; G06N3/04 ; G06K9/62
摘要:
本发明公开了一种面向轻量级卷积神经网络的权值数值定点量化方法,本发明通过对权重划分,再通过特定的集合中提取量化策略,进行训练,完成多测迭代,得到量化因子,完成量化。本发明相比于现有技术,具有更快的再训练速度,在轻量级卷积神经网络上较低的准确率损失,实现了比卷积8位整数乘法化更低的硬件资源消耗。
摘要(英):
The invention discloses a weight value fixed-point quantization method for a lightweight convolutional neural network. The method comprises the steps of dividing weights, extracting a quantization strategy from a specific set, training, completing multi-measurement iteration, obtaining a quantization factor and completing quantization. Compared with the prior art, the method has the advantages that the retraining speed is higher, the accuracy loss is lower on the lightweight convolutional neural network, and the hardware resource consumption is lower than that of convolutional 8-bit integer multiplication.