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
- 暑假第一周
- Halcon shape_ trans
- Guess riddles (9)
- RT thread kernel quick start, kernel implementation and application development learning with notes
- Guess riddles (11)
- Guess riddles (7)
- Program error record 1:valueerror: invalid literal for int() with base 10: '2.3‘
- Guess riddles (8)
- Numpy 小坑:维度 (n, 1) 和 维度 (n, ) 数组相加运算后维度变为 (n, n)
- [Niuke brush questions day4] jz55 depth of binary tree
猜你喜欢

Beautiful soup parsing and extracting data

RT thread kernel quick start, kernel implementation and application development learning with notes

319. Bulb switch

How to manage the performance of R & D team?

Mathematical modeling: factor analysis

Digital analog 1: linear programming
![[Niuke brush questions day4] jz55 depth of binary tree](/img/f7/ca8ad43b8d9bf13df949b2f00f6d6c.png)
[Niuke brush questions day4] jz55 depth of binary tree

Add discount recharge and discount shadow ticket plug-ins to the resource realization applet

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
随机推荐
File server migration scheme of a company
How many checks does kubedm series-01-preflight have
Count of C # LINQ source code analysis
Redis实现高性能的全文搜索引擎---RediSearch
Meta tag details
LLVM之父Chris Lattner:为什么我们要重建AI基础设施软件
Characteristic Engineering
Search data in geo database
Tips 1: Web video playback code
Guess riddles (8)
Ros-10 roslaunch summary
Redis implements a high-performance full-text search engine -- redisearch
Mengxin summary of LCs (longest identical subsequence) topics
Basic number theory -- Euler function
优先级队列(堆)
暑假第一周
Basic number theory - fast power
AUTOSAR从入门到精通100讲(103)-dbc文件的格式以及创建详解
location search 属性获取登录用户名
It cold knowledge (updating ing~)