解决现实世界问题的四大机器学习策略

曾经终败给现在 2022-04-18 02:39 164阅读 0赞

有的在线学习算法还会“适应”，即随时调整模型，跟随或追踪数据的变化。具体而言，它们会逐渐“遗忘”过时数据，从而适用于环境/模型随时变动的应用程序。

**原文翻译：**

机器学习有四种广受认可的形式：监督式、无监督式、半监督式和强化式。在研究文献中，这些形式得到了深入的探讨。它们也被纳入了大多数机器学习算法的入门课程。下表对这四种形式作了总结。

<table> 
 <thead> 
 <tr> 
 <td> Style of Learning </td> 
 <td> Definition </td> 
 <td> Characteristic </td> 
 <td> Examples </td> 
 </tr> 
 </thead> 
 <tbody> 
 <tr> 
 <td> Supervised </td> 
 <td> Supervised learning algorithms make predictions based on a set of examples. </td> 
 <td> Data has known labels. </td> 
 <td> Insurance underwriting and fraud </td> 
 </tr> 
 <tr> 
 <td> Unsupervised </td> 
 <td> With unsupervised learning, there is a focus on finding patterns and gaining insight from data. </td> 
 <td> Labels or output are unknown. </td> 
 <td> Customer clustering and market basket analysis </td> 
 </tr> 
 <tr> 
 <td> Semi-supervised </td> 
 <td> With semi-supervised learning, you use unlabeled examples with a small amount of labeled data to improve the learning accuracy. </td> 
 <td> Labels or output are known for a subset of data. </td> 
 <td> Medical predictions where tests and expert diagnoses are expensive and only part of the population receives them. </td> 
 </tr> 
 <tr> 
 <td> Reinforcement </td> 
 <td> Reinforcement learning analyzes and optimizes the behavior of an agent based on the feedback from the environment. </td> 
 <td> Focus on making decisions based on previous experience. </td> 
 <td> Robotics, gaming, navigation, complex decision problems and reward systems. </td> 
 </tr> 
 </tbody> 
</table>

但一个并不广为人知的概念是机器学习策略，即以创造性的方式，运用基本的机器学习算法，解决现实世界问题。我对这些策略产生了浓厚的兴趣。在这篇文章中，我将探讨四种策略：在线学习、迁移学习、集成学习和深度学习。好在，这四项策略适用于表格中任意一种机器学习形式。

[原文链接][Link 1]

[Link 1]: http://click.aliyun.com/m/1000022479/