How to classify unsupervised learning

1 Unsupervised learning

Labeled is supervised learning, unlabeled is unsupervised and of course there is semi-supervised learning

Unsupervised learning algorithms can split this data into two different clusters. So-called clustering algorithm

Unsupervised learning automatically classifies data

Give it some fruits, it can classify which are bananas and which are apple durian. Automatic classification


2 application

This includes understanding and applying genetics. An example of DNA microdata. The basic idea is to enter a group of different individuals and for each of them it must be analyzed whether they have a particular gene. Technically it has to be analyzed how many specific genes have been expressed. These colors, red, green, gray, etc., indicate the appropriate degree, that is, whether different individuals have a particular gene. You can run a clustering algorithm to group individuals into different classes or different types of groups (people) ...

So this is unsupervised learning as we did not give the algorithm some information in advance, e.g. B. This is the first type of person, these are the second type and the third type and so on. We just said yes, that's a lot of data. I don't know what's in the data. I don't know who and what type is. I don't even know what types of people there are and what types. But can you find the structure in the data automatically? It means you want to automatically group these people into different classes, and I can't know in advance which are which. Since we didn't give the algorithm the correct answer in response to the data in the dataset, this is unsupervised learning


Unsupervised learning or aggregation has a large number of uses. It is used to organize large computer clusters. Some of my friends work in large data centers that have large computer clusters. You want to find out what types of machines are easy to work with. When you let these machines work together, you can make your data center run more efficiently. The second application is the analysis of social networks. So if you know your friends' information, such as: E.g. those you email frequently or your Facebook friends, Google+ circle friends, can we group friends automatically? That means the people in each group know each other well, do everyone in the group know? There is also market segmentation. Many companies have large databases that store consumer information. Hence, you can get these customer records, automatically detect market categories, and automatically divide customers into different market segments so that you can automatically and more effectively sell in different market segments or sell together. This is also unsupervised learning, as we all have customer data but don't know in advance which market segments and which customers are in our data set. We don't know who is in the first market segment, who is in the second market, and so on. Then we have to let the algorithm figure all of this out of the data. Finally, unsupervised learning can also be used for astronomical data analysis. These clustering algorithms provide surprising, interesting, and useful theories to explain how galaxies are born. These are all examples of clustering, which is just a type of unsupervised learning


I'll tell you another one now. First, let me introduce the topic of cocktail party. Did you go to a cocktail party? You can imagine there is a banquet hall full of people all sitting and chatting, so many people chatting at the same time and their voices overlapping because everyone is talking, talking at the same time, which you can barely hear the voice of the person in front of you . So maybe two people at a cocktail party like this, both talking at the same time, let's say they're at a little cocktail party now. We put two microphones in the room because these microphones are in two places and the distance to the speaker is different. Each microphone records different sounds even though they are the same two speakers. It sounds like two recordings are being superimposed or summed together to produce the recordings we have now. In addition, this algorithm also distinguishes between two audio resources that can be synthesized or merged with the previous recording. In fact, the first output result of the Cocktail algorithm is:


Therefore the English voice was separated from the recording.

The second edition is as follows:



Take a look at this unsupervised learning algorithm, how complicated it is to implement, right? It appears to be so. To build this application, you will need to write a lot of code to complete this audio processing or link to a number of JAVA synthesizer libraries. The audio processing library looks absolutely complicated. The audio is separated from the audio. In fact, the algorithm that corresponds to the problem just known can be carried out with just one line of code.

[W, s, v] = svd ((repmat (sum (x. * X, 1), size (x, 1), 1). * X) * x ');


Reprinted at: