The book "Machine Learning" is a good introductory book for understanding machine learning knowledge. It is a masterpiece written by Mr. Zhou Zhihua from Nanjing University. I heard about Mr. Zhou's name a long time ago. It is considered a domestic machine learning field. There are a few big cows. It happens that the graduate students are doing the content related to this direction, so I bought a so-called watermelon book today, and I am going to study it. I am not good at memorizing just reading, and I am deeply impressed by taking notes and practicing while reading. Your own learning process is organized as follows according to the content of each chapter:
Day1 Chapter 1 Introduction
The author of this book, Mr. Zhou Zhihua, begins the chapter by chatting, introducing to readers what is machine learning and what is learning algorithm with small cases and scenarios of daily life. This chapter introduces a lot of machine learning related terminology concepts.
First of all, to learn, you must first have data. The collection of object records we want to learn is called a "data set", and the description of the records and objects in it is called an "instance". Or "sample", a matter that reflects the performance or nature of events or objects in a collection, we call it "attribute" or "feature", and the space formed by attributes is called "attribute space", "sample space" or "input space", since each point in the space corresponds to a coordinate vector, we call an example a "feature vector" ", where the number of attributes is what we call the "dimensionality" of the sample
The data is obtained above. The process of learning the model from the data is called "learning" or "training". In this process, a certain learning algorithm is executed to complete the process. The data used in the training process is called "learning" or "training". Training data (training data)", each of which is called a "training sample", and the set composed of the training samples is called "training set", and the model obtained by training corresponds to A certain underlying law of the data, this result is called "hypothesis", we use the learned result to "predict", and the process of predicting with the learned model is called "testing" ”, the predicted sample is called “testing sample”.
What we want to predict is discrete values. This kind of learning task is called "classification". What we want to predict is continuous value. We call this kind of learning task "regression". Of course, we can also do "regression" on the data. Clustering", that is, dividing the objects in the training set into several groups, each group is called a "cluster".
According to whether the training data has label information, learning tasks can be roughly divided into two categories: "supervised learning" and "unsupervised learning". Classification and regression are the representatives of the former, and clustering is the latter. representative of the person.
The ability of the learned model to apply to new examples is called "generation".
Usually, we assume that all samples in the sample space obey an unknown "distribution", and each sample we obtain is independently sampled from this distribution, that is, "independent and identically distributed (iid)" .
Induction and deduction are the two basic means of scientific reasoning. The former is a process of "generation" from special to general, and the latter is a process of "specialization" from general to special. For example, in mathematics, a consistent theorem is derived from mathematical axioms, which is a deductive process, and "learning from samples" is an inductive process, called "inductive learning".
There is inductive preference in inductive learning, which follows the principle of Occam's razor.
Development history: Machine learning is an inevitable product of artificial intelligence research to a certain stage. From the 1950s to the 1970s, artificial intelligence research was in the "inference period". At that time, people believed that as long as the machine could be given the ability to logically reason, the machine could have intelligence. Since the mid-1970s, artificial intelligence research has entered the "knowledge period", during which a large number of expert systems have come out. The 1980s was a period when machine learning became an independent subject field, and various machine learning technologies were blooming. In the mid-1990s, "statistical learning" debuted and quickly took over the mainstream stage, represented by the support vector machine (SVM) and the more general "kernel methods". ". At the beginning of the 21st century, connectionist learning came back again ("connectionism" based on neural networks in the mid-to-late 1950s), setting off an upsurge in the name of "deep learning". many layers" neural network
Now, machine learning has developed into a fairly large subject area. Today's increase in computing power and the blessing of big data have gradually pushed machine learning to a climax.
(The first chapter notes are here, continue to study the subsequent chapters)