Simple explanation of "difference between machine learning and deep learning" for beginners
"AI (artificial intelligence) is a big topic in the world of computer science," Deep learning "is supporting that technology. On the other hand, we often hear the word "machine learning" using computers, but in fact there are many people who do not understand the difference well. The difference between these two is the developer of mathematical calculation software "MATLAB"MathWorksIt explains briefly.
Introduction to Deep Learning: Machine Learning vs Deep Learning - YouTube
Machine learning and deep learning are techniques used to classify data by providing learning models. An example of classifying images of dogs and cats is often used to explain its function. For this image, almost everyone should answer that the left is a dog and the right is a cat.
However, when you bring another image, you can determine whether it is a cat or a dog by analyzing the image in the brain, seeing its features, and judging the ability to judge according to knowledge and experience Because it is equipped.
It is the purpose of machine learning and deep learning that let the computer process this same thing. However, there is a difference in approach between the two.
In the case of machine learning, humans preprocess materials first. In the case of image, by keeping it indicates that it is a corner or boundary lines, computer so that you can recognize. Then, the computer leave accumulated by analyzing the features that are included in the data, to derive the answer from the last image of the features the accumulated data to the original.
The example above was an example of object recognition, but the same method can be used for scene recognition and object detection.
That is, the workflow of machine learning is as follows. First of all, there is an image, learning is made by letting the machine (= computer) recognize its features, and solving the last given problem.
In the case of deep learning, the work which human beings do is omitted. If the original data is given as it is, the deep learning algorithmConvolution neural networkWe can analyze ourselves by making full use of it and get answers.
The It must be recognized here is actually "deep learning" are intended to be included within the concept of machine learning in a broad sense, and say fact. In other words it such means that "machine learning> deep learning", but in this movie describes the machine learning in a manner of classification of "things that are not deep learning".
So, how to use machine learning and deep learning properly? This can be judged appropriately by "hardware performance" and "amount of data". Since deep learning performs enormous computation processing on enormous amount of data, hardware such as GPU with high computing power is essential. Therefore, if you are blessed with hardware, if there are more data, deep learning, the performance of the machine is not so high, and if the number of data to be processed is not so much, it is good to adopt machine learning.
If you choose machine learning, you have the option of using a large number of classifiers when learning. It is also possible to choose which feature to use to guide optimum answers. In the case of machine learning, there is an advantage that it can be selected by combining approaches according to purpose.
In the case of machine learning, there is still room for inclusion of our ingenuity, and the time required for processing is short, whereas the feature that deep learning can obtain high precision even if the user does not have much knowledge or ingenuity There is. However, in case of deep learning, there is a disadvantage that high machine performance and long processing time take. Also, in order to learn with deep learning with high precision, enormous amounts of data are required, and internal algorithms are "black boxes" that humans can not understand any longer, so it is virtually impossible to debug It also needs to be remembered.
In other words, machine learning and deep learning are strictly included in the same "machine learning", and their choice should be selected according to the amount of data to be processed, the performance of the hardware, and the contents of the answer to be obtained That is why.