2018年08月_春江花月夜晨

12月 08月 05月 04月 03月 02月

原创 Rotating Data for Neural Network Computations

Fig.11 例1100展示利用二维脉动阵列1114如何旋转kernel矩阵结构用于计算卷积运算。输入为1102，将feature map tile成5x5的matrix，weight为1104，是3x3的kernel，第一个layer的feature map为170x170。 Inputs的size是由kernel大小决定的，tile_x=2*kx-2，tile_y=2*ky-...

2018-08-17 16:45:55 366

原创论文DianNao总结(重点在后半段) A Small-Footprint High-Throughput Accelerator for Ubiquitous Machine-Learning

参考论文： 1、https://blog.csdn.net/u013108511/article/details/78577405 2、https://yiweifen.com/v-1-337993.html 3、https://blog.csdn.net/evolone/article/details/80765094 4、http://baijiahao.baidu.com/s...

2018-08-15 17:13:26 5046 1

need_to_do.bc

需要处理的3d_rendering含有大量memcpy，各种不同情况的src和dest，主要原因就是9个i8，会内存拼接为i64和i8

2020-03-11

computing convolutions using a neural network processor

2018-12-14

batch processing in a neural network processor

google batch processing in a neural network processor

2018-12-14

rotating data for neural network computations

google rotating data for neural network computations

2018-12-14

vector computation unit in a neural network processor

2018-12-14

COMPRESSING DEEP NEURAL.pdf Efficient Speech Recognition Engine.pdf EIE Efficient Inference Engine on Compressed Deep Neural Network.pdf Exploring the Regularity of Sparse Structure in Convolutional Neural Networks

2018-12-12

寒武纪的DaDianNao

这是寒武纪的第二篇论文DaDianNao，google上下载的原文

2018-07-17

对caffe的训练model和prototxt文件进行layer参数读取，主要对conv、pool和inner里面参数读取

2018-04-17

空空如也

TA创建的收藏夹 TA关注的收藏夹

TA关注的人

春江花月夜

原创 Rotating Data for Neural Network Computations

原创论文DianNao总结(重点在后半段) A Small-Footprint High-Throughput Accelerator for Ubiquitous Machine-Learning

llvm.ppt llvm使用及其组成简介

need_to_do.bc

computing convolutions using a neural network processor

batch processing in a neural network processor

rotating data for neural network computations

vector computation unit in a neural network processor

深鉴科技、韩松的相关论文

寒武纪的DaDianNao

对caffe的训练model和prototxt文件进行layer参数读取，主要对conv、pool和inner里面参数读取

空空如也

原创 Rotating Data for Neural Network Computations

原创 论文DianNao总结(重点在后半段) A Small-Footprint High-Throughput Accelerator for Ubiquitous Machine-Learning

llvm.ppt llvm使用及其组成简介

need_to_do.bc

computing convolutions using a neural network processor

batch processing in a neural network processor

rotating data for neural network computations

vector computation unit in a neural network processor

深鉴科技、韩松的相关论文

寒武纪的DaDianNao

对caffe的训练model和prototxt文件进行layer参数读取，主要对conv、pool和inner里面参数读取

空空如也

原创论文DianNao总结(重点在后半段) A Small-Footprint High-Throughput Accelerator for Ubiquitous Machine-Learning