plasmascience的个人博客分享 http://blog.sciencenet.cn/u/plasmascience

博文

机器学习

已有 1541 次阅读 2022-5-20 10:55 |系统分类:科研笔记

各行各业都在试机器学习,磁约束核聚变也不例外。

最近以磁平衡重构为切人点,尝试了机器学习中最重要的一种方法--神经网络。


费米曾批评一个含有很多自由参数的模型说:我的朋友诺伊曼说,如果给我四个参数,我可以拟合出一头大象,如果增加到5个参数,我可以让大象挥舞鼻子。


费米这里隐含的一个意思是, 拟合已有的数据是很容易的,但这样的模型没有预测能力,他认为基于物理原理的模型才有可靠的预测能力,而基于数据拟合的经验模型预测能力不可靠。这也是物理学家普遍持有的一种观点。

能不能增加自由参数的数量,让拟合得到的模型有预测能力呢?

人工神经网路回答了这一问题:可以。 但自由参数的数量提高到千万量级,90年代以前训练这种网络(找到最优的拟合参数)很困难。

1986年反向传播算法被发现,极大地提高了这种网络的训练效率,神经网路迅速找到很多应用场景,解决了很多传统人工智能无法解决的问题。 



我的机器学习笔记: machine_learning.pdf


Enrico Fermi once criticized the complexity of a model (that contains many free parameters) by quoting Johnny von Neumann “With four parameters I can fit an elephant, and with five I can make him wiggle his trunk”.


What Fermi implies is that it is easy to fit existing data and what is importan is to have a model with predicting capability (fitting data not seen yet). The artificial neural network method tackles this difficulty by increasing the number of free parameters to millions, in the hope of obtaining predicting capability.




https://blog.sciencenet.cn/blog-620148-1339409.html

上一篇:越想越复杂,越做越简单
下一篇:简单说说麦克斯韦方程
收藏 IP: 218.106.82.*| 热度|

1 李振乾

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2024-4-23 16:40

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部