Blog of YHR
Blog of YHR
Konck! Knock!
OK
Blog of YHR
Choose mode
dark
auto
light
Home
Category
AIDistributedTraining
AIGC
Distributed Transaction
Distributed Lock
Algorithm
Android
Architecture
Cloud
DevOps
Computer_network
VHDL
Data structure
Data warehouse
DL
Distributed System
Docker
ElasticSearch
Flink
Go
Git
HuaweiCloud
Haskell
NLP
JS
Java
JavaWeb
JVM
Jenkins
Kafka
Kubernetes
LLMs
Leetcode
MDSD
Linux
MongoDB
DB
Nginx
Nodejs
Notes
OS
OpenStack
PHP
Python
RabbitMQ
Prometheus
React
RealTimeSE
Redis
Spring Cloud
Service Computing
Spring
Springboot
SpringMVC
Spring Security
Swagger
SuperSet
Tensorflow
Vuepress
XML
Zookeeper
K8s
Markdown
Mybatis
Solr
Tag
TimeLine
Contact
GitHub
(opens new window)
Ch13 CNN
Konck! Knock!
OK
Ch13 CNN
Yang Haoran
7/1/2022
Tensorflow
CNN
#
Ch13 CNN
#
卷积概念
#
感受野
计算量:(x-2)²
9 + (x-2-2)²
9
#
使用padding让输入与输出大小一样
#
TF描述卷积层
#
Batch normalization
神经网络对零附近的值比较敏感,所以用BN使把偏移的数据重新拉回
但是通常这样会把数据拉回‘线性区域‘,也就是0附近
有可能会丧失非线性特征,所以引入两个可训练参数,使得数据可以缩放和位移,保证非线性特征
用在卷积层和激活层之间
#
Pooling池化
#
Dropout舍弃
#
CNN
卷积概念
感受野
使用padding让输入与输出大小一样
TF描述卷积层
Batch normalization
Pooling池化
Dropout舍弃
CNN