没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
Understanding Learning Rates and How It Improves Performance in Deep Learning This post is an attempt to document my understanding on the following topic: What is the learning rate? What is it’s signibcance? How does one systematically arrive at a good learning rate? Why do we change the learning rate during training? How do we deal with learning rates when using pretrained model?
资源推荐
资源详情
资源评论
HadzZulkii
Follow
DataScientistatSEEK
Jan22 · 8minread
UnderstandingLearningRatesandHowIt
ImprovesPerformanceinDeepLearning
This post is an attempt to document my understanding on the following
topic:
What is the learning rate? What is it’s signicance?
How does one systematically arrive at a good learning rate?
Why do we change the learning rate during training?
How do we deal with learning rates when using pretrained model?
Much of this post are based on the stu written by past fast.ai fellows
[1], [2], [5] and [3]. This is a concise version of it, arranged in a way
for one to quickly get to the meat of the material. Do go over the
references for more details.
Firsto,whatisalearningrate?
Learning rate is a hyper-parameter that controls how much we are
adjusting the weights of our network with respect the loss gradient. The
lower the value, the slower we travel along the downward slope. While
this might be a good idea (using a low learning rate) in terms of making
sure that we do not miss any local minima, it could also mean that we’ll
be taking a long time to converge — especially if we get stuck on a
plateau region.
The following formula shows the relationship.
new_weight = existing_weight — learning_rate * gradient
•
•
•
•
Typically learning rates are congured naively at random by the user.
At best, the user would leverage on past experiences (or other types of
learning material) to gain the intuition on what is the best value to use
in setting learning rates.
As such, it’s often hard to get it right. The below diagram demonstrates
the dierent scenarios one can fall into when conguring the learning
rate.
Gradientdescentwithsmall(top)andlarge(bottom)learningrates.Source:AndrewNg’sMachine
LearningcourseonCoursera
Eectofvariouslearningratesonconvergence(ImgCredit:cs231n)
剩余11页未读,继续阅读
资源评论
leez888
- 粉丝: 0
- 资源: 2
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- C#/WinForm演示退火算法(源码)
- 如何在 IntelliJ IDEA 中去掉 Java 方法注释后的空行.md
- 小程序官方组件库,内含各种组件实例,以及调用方式,多种UI可修改
- 2011年URL缩短服务JSON数据集
- Kaggle-Pokemon with stats(宠物小精灵数据)
- Harbor 最新v2.12.0的ARM64版离线安装包
- 【VUE网站静态模板】Uniapp 框架开发响应式网站,企业项目官网-APP,web网站,小程序快速生成 多语言:支持中文简体,中文繁体,英语
- 使用哈夫曼编码来对字符串进行编码HuffmanEncodingExample
- Ti芯片C2000内核手册
- c语言实现的花式爱心源码
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功