1.1 Introduction
I was most impressed by the words: machine classification ability than people.
Why should we learn Machine Vision? Professor Zhou Zhihua begins with several properties watermelon (such as color, pedicle, knocking sound) to determine whether a watermelon is good melon, these are done by human experience.
This means one door machine learning discipline dedicated to the study by means of a computer, use the experience to improve their system performance. The main contents of machine learning: learning about "learning algorithm". With the experience of our learning algorithm data to him, he will be able to produce a corresponding model in the face of the new situation, the model will provide us with a corresponding judgment.
The relationship between machine learning and data mining: a lot of people began to be data mining, ah, ah Liaode large data dazzling, in fact, the essence of machine learning is the king. Directly on the map:
The term substantially 1.2
Data Set: is a combination of a set of records, wherein the training data used in the process is called the training data, the test is called test set, to verify and to achieve the parameter adjustment purpose are the validation set, the test set is different, the test set main surface the performance and generalization ability of the test model.
Properties: or called features, or to reflect events or performance properties of an object in a certain area.
Properties Space: property as a multi-dimensional spatial coordinates composition. As a watermelon Zhang space: three-dimensional coordinate axes knocking sound, pedicle color thereof. This is where particular emphasis on conceptual understanding of space, matrix theory interpretation of space it is in place.
Category: Predicting goal is to learn the task of discrete values are classified, called a binary only two categories.
Return: Learning is a continuous task to predict a target value is a return.
Clustering: The target data set into a plurality of disjoint clusters sample.
Generalization: the ability to learn new model is applicable to samples.
Hypothesis space: the learning process to see the process of searching in a space as consisting of all the assumptions, the goal is to find and search training set "match" assumption; in this hypothesis space, and there may be multiple hypotheses consistent training set, we call it "version of the space."
Summarized preference: machine learning algorithms preferences for certain types of assumptions in the learning process. Or simply preferences such as: morning review of mathematics, English is the afternoon than the morning review review review English afternoon mathematics stronger. Of course, there is no free lunch, when you use some kind of summed up the current application preferences may be better for you, but might be missing in the other.
Occam's razor: is a common, natural science research in the basic principle that "if more than one observation is consistent with the hypothesis, that the easiest option."
对于没有免费的午餐，我们来看下下面的讨论：
可以看出无论算法有多聪明，最后的期望性能竟然相同，但是别慌：NFL免费的午餐定理的前提条件是所有问题同等重要，所以说什么都要联系实际，实践是真理的唯一标准。
1.3发展历程
1.4应用现状
1.5阅读材料
