What is deep learning? Easy-to-understand explanation of differences and mechanisms from machine learning and practical examples
Deep learning is one of the technologies used in the AI field. This is a machine learning method that uses a multi-layered neural network. By ensuring a sufficient amount of data and training it, AI will be able to automatically extract features from the data.
Deep learning has made it possible to learn unstructured data (images, natural language, sound) that was difficult to digitize in the past, and it can now be used for image recognition, voice recognition, natural language processing, and abnormality detection. rice field.
In this article, the meaning and mechanism of deep learning, typical algorithms, usage methods, practical examples, etc. are explained in an easy-to-understand manner.
What is deep learning?
I will explain the meaning and mechanism of the words of deep learning.
Meaning of deep learning
Deep learning is a learning method that utilizes neural networks, which are typical algorithms for machine learning.
★ What is a neural network?
A neural network is an AI that models the structure and function of neurons (nerve cells that make up the brain of an organism).
A neural network consists of an input layer that stores data, an intermediate layer (hidden layer) that processes weights flowing from the input layer, and an output layer that outputs results.
How deep learning works
In deep learning, the neural network itself can automatically extract the characteristics of the data group as long as there is sufficient learning data.
Since the multi-scale intermediate layer cuts the input data into various sizes and determines the characteristics, it extracts from the detailed pattern to the large structure and the entire outline based on the given data.
We are good at pattern recognition of data that cannot be symbolized such as images.
Difference from machine learning
Machine learning is "learning" in AI. It has the meaning of "learning by the machine itself" just as humans learn.
Machine learning is one of the technologies that support AI, and deep learning is one of the methods of machine learning.
A typical deep learning algorithm
Here are some typical deep learning methods.
CNN (Convolutional Neural Network)
CNN (Convolutional Neural Network) is mainly used for image recognition and motion detection. It consists of a "convolution layer" that extracts the image's features and a "pooling layer" that analyzes the attributes. It has a high ability to recognize patterns for images and is characterized by its ability to recognize quickly.
RNN (Recurrent Neural Network)
RNN (Recurrent Neural Network) is a general term for neural networks with an autoregressive architecture, and it can handle time series data of variable length. It is mainly used in voice recognition, video recognition, natural language processing, etc.
How to use deep learning
Deep learning made it possible for AI to learn unstructured data (images, natural language, and sounds) that were previously difficult to digitize.
The increasing variation in digitization is improving the accuracy of optimization and recommendations. Currently, it is used in various situations such as image recognition and voice recognition.
The main uses of deep learning are as follows.
How to use ①Image recognition
Technology that recognizes faces and human figures from photos. The feature separates from the background of the input image or video, and the target target feature is extracted.
Example: face recognition on iPhone, tagging on Facebook
← Example of Business Efficiency Improvement Using AI Image Recognition
How to use ②Voice Recognition
Technology that recognizes the human voice. You can recognize people by voice or voice input.
Example: Siri and Alexa voice input
Usage: natural language processing
Technology that makes computers understand written and spoken words used in everyday communication.
Examples: machine translation, language modeling, answering questions
Usage ④ Abnormality Detection
A technology that detects anomalies using time series data collected from sensors.
Example: credit card fraud and manufacturing quality control
Now, the technology of interest is "GAN (Anti-Generation Network)"
Among the technologies that use deep learning, “Generative Adversarial Networks (GAN)” attract particular attention.
GAN is a kind of generative model and it consists of two networks, namely a generative network and a limiting network. The main advantage is that the generator learns to deceive the discriminator, and the discriminator learns to distinguish more accurately. The name "hostile" is used because the two networks are learning for conflicting purposes.
It is possible to create data that does not exist when creating an image, or transform according to the properties of the existing data. For example, if you use a GAN-based tool called "IMAGE IN PAINTING" published by NVIDIA, you can scan people and objects from images and process them only in the background.
As an example, if the purpose is to generate an image, the generation side outputs an image, and the identification side judges whether the image is correct or not. The generator learns to deceive the discriminator, and the discriminator learns to discriminate more accurately. It is called hostile because the two networks learn for conflicting purposes in this way.