Machine learning is one of the hottest words in recent years, just like “big data”. Before we understand deep learning, it is necessary to understand and recognize the term machine learning. What exactly is machine learning a process or behavior?
Machine learning — what is the purpose of human learning? It’s knowledge, it’s ability, it’s skill, and eventually it’s able to do relatively complex or demanding work. So if you think about a machine, we have a machine learn, whatever it learns, the ultimate goal is to have it independently or at least semi-independently do relatively complex or demanding work. What do we do
The machine learning mentioned here is to let machines help humans do some large-scale data recognition, sorting, rule summary and other things that take time for humans to do, which is the essential purpose of machine learning.
In the long history of human development, machine gradually replace human production work is an irreversible trend of slash-and-burn agriculture from the primitive, a large number of the primitive clan, tribe is primitive tools used are hunting together, then with the continuous phase of productivity development and the type of work divide five stimulation, more and more tools appear, instead of the original production need
Lots of people.
In modern times, especially after the first and second industrial Revolutions, high-energy machines powered by fossil fuels once again replaced human and animal power in more fields, greatly improving human productivity.
After the information revolution, along with the computer calculation, and a new theory in the field of computer algorithms the gradual development of new technology, machines also gradually replace man, to participate in more information with “a certain intelligent” sorting and identification of the work There we important mention the “smart”.
Algorithm for this kind of things in the presence of the original is a deterministic machine instruction execution sequence, that is to say, the machine need to do is set from the start as early as the program while in the process of program execution can rely on limited parameters involved in the process of program execution object, execution times, executive branch conditions, such as setting, but the basic behavior logic
Series has been generally identified In this process, the machine one by one computer is very passive, it honestly strict execution sequence of instructions that programmers give it, there is no “learning” This didn’t also way, because the beginning of the Turing machine model at the time of design is expect computers to work this way.
And the field of machine learning ideas, and we have just told me that there is a big difference in a form We supervised learning process, for example, such as a classifier, can be the input mail sorting for “ordinary mail” and “spam” two categories But for spam judgment standard is not given at the beginning of the program is run, After a large amount of junk mail is given to the classifier, the classifier calculates and summarizes the various characteristics of the junk mail sample, and then gets it.
In this training process, a large number of messages labeled as spam are given to the classifier, which is called training sample. (The process of classifying and summarizing the characteristics of spam samples is called training. (The judgment criteria summarized by Traning are called classification model. We’ll also pick up other “regular mail” and
“Spam” is given to the classifier to try to classify according to the classification model just summarized to see whether its classification correctness is really generalization. Validation is mainly used to verify whether the data relationships summarized from the training samples can be migrated. After that, quantitative “ordinary mail” and “spam” are used to test the recognition ability of the model. Let’s see if it’s within the scope that our business allows. This is basically a complete supervised learning process.
Other supervised learning scenarios are similarly based on training samples and validation using validated data sets. In the process, we see not hard, finally the whole classifier tool for people producing environmental service way of working is not my prior written in the program, but to give the first samples show the classification rules (standard), then classified according to the rules of this operation This is a very image of the “machine learning” process, Isn’t it interesting that in the process it learns on its own how to classify things correctly?
Machine learning in terms of the types of learning, the most common machine learning can be divided into two types: one is unsupervised learning and the other is supervised learning. It refers to the kind of mining work in which people try to find out the internal implication relationship without labels after obtaining the trained vector data. In this process, users do not need to do any marking or even process intervention on these samples except for setting some necessary hyper-parameters. Supervised learning is different, each sample has a specific label, and at the end of the day we’re just going to figure out the mapping between the vector of the training sample and the label. So there’s a big difference in logic between these two approaches, and you need to be very careful if you’re new to this.
This article is excerpted from Vernacular Deep Learning and TensorFlow. Click here for a video tutorial on deep learning and ARTIFICIAL intelligence.