当前位置:网站首页>Huber Loss
Huber Loss
2022-07-05 08:51:00 【Wanderer001】
Huber Loss 是一个用于回归问题的带参损失函数, 优点是能增强平方误差损失函数(MSE, mean square error)对离群点的鲁棒性。
当预测偏差小于 δ 时,它采用平方误差,当预测偏差大于 δ 时,采用的线性误差。
相比于最小二乘的线性回归,HuberLoss降低了对离群点的惩罚程度,所以 HuberLoss 是一种常用的鲁棒的回归损失函数。
Huber Loss 定义如下

参数 a 通常表示 residuals,写作 y−f(x),当 a = y−f(x) 时,Huber loss 定义为:

δ 是 HuberLoss 的参数,y是真实值,f(x)是模型的预测值, 且由定义可知 Huber Loss 处处可导。
边栏推荐
- Warning: retrying occurs during PIP installation
- [daiy4] copy of JZ35 complex linked list
- Halcon snap, get the area and position of coins
- Reasons for the insecurity of C language standard function scanf
- Codeworks round 638 (Div. 2) cute new problem solution
- TypeScript手把手教程,简单易懂
- 我从技术到产品经理的几点体会
- Business modeling of software model | stakeholders
- Multiple linear regression (gradient descent method)
- AUTOSAR从入门到精通100讲(103)-dbc文件的格式以及创建详解
猜你喜欢

Redis implements a high-performance full-text search engine -- redisearch

It cold knowledge (updating ing~)

EA introduction notes

Business modeling of software model | vision

Numpy pit: after the addition of dimension (n, 1) and dimension (n,) array, the dimension becomes (n, n)

Classification of plastic surgery: short in long long long

深度学习模型与湿实验的结合,有望用于代谢通量分析

Halcon clolor_ pieces. Hedv: classifier_ Color recognition

TF coordinate transformation of common components of ros-9 ROS

Install the CPU version of tensorflow+cuda+cudnn (ultra detailed)
随机推荐
notepad++
微信H5公众号获取openid爬坑记
Guess riddles (2)
Codeworks round 638 (Div. 2) cute new problem solution
golang 基础 —— golang 向 mysql 插入的时间数据和本地时间不一致
12、动态链接库,dll
The location search property gets the login user name
Infix expression evaluation
Typescript hands-on tutorial, easy to understand
GEO数据库中搜索数据
多元线性回归(梯度下降法)
kubeadm系列-00-overview
Solutions of ordinary differential equations (2) examples
It cold knowledge (updating ing~)
牛顿迭代法(解非线性方程)
AdaBoost use
js异步错误处理
uni-app 实现全局变量
Illustration of eight classic pointer written test questions
[daiy4] jz32 print binary tree from top to bottom