Skip to content

Chapter 5 Machine Learning Basics

每章的开头作者都会抛出一系列的问题，本章作者所抛出的问题可以说是machine learning中最最本质的问题：

  • What a learning algorithm is
  • How the challenge of fitting the training data differs from the challenge of finding patterns that generalize to new data
  • How to set hyperparameters

Machine learning is essentially a form of applied statistics with increased emphasis on the use of computers to statistically estimate complicated functions and a decreased emphasis on proving confidence intervals around these functions; we therefore present the two central approaches to statistics: frequentist estimators and Bayesian inference.

这段话如何理解？

本质上来说，“machine learning”属“applied statistics”。所以要理解上面这段话需要对Statistics的研究分支有一些了解了。上面这段话中的“frequentist estimators”和“Bayesian inference”都属于Statistical inference范轴，按照Statistical inference的说法，“frequentist estimators ”对应的是Frequentist inference，“Bayesian inference”对应的是Bayesian inference，它们是Frequentist inference的两个学派（或paradigm），关于Frequentist inference的学派，参见Paradigms for inference。需要注意的是：

These schools—or "paradigms"—are not mutually exclusive, and methods that work well under one paradigm often have attractive interpretations under other paradigms.

“confidence intervals”的中文意思是：置信区间，参见Confidence Interval。

关于统计学（statistics）的内容，参见Probability-theory-and-Statistics章节。

We describe how to combine various algorithm components such as an optimization algorithm, a cost function, a model, and a dataset to build a machine learning algorithm.

我觉得这段话所传达的思想是比较好的：如果将machine learning algorithm看做是一个机器的，那么它有如下零件（component）组成：

  • optimization algorithm
  • model
  • dataset

那读者看到会提出这样的问题：

  • 这些component分别表示的是什么？
  • 它们之间是如何组装、协助来构成一个完整的machine learning algorithm？
  • 每个component有哪些可供选择的option？

这些问题在本书的后续章节会专门进行介绍。

Finally, in section , we describe some of the 5.11 factors that have limited the ability of traditional machine learning to generalize. These challenges have motivated the development of deep learning algorithms that overcome these obstacles.

上述ability的含义是什么？deep learning algorithm较traditional machine learning algorithm的优势何在？