Deep learning, machine learning, neural networks, what is the relationship between them?

For a long time, the heat of artificial intelligence has been maintained at a certain height. But when you are paying attention to or studying the field of artificial intelligence, you will always encounter such keywords: deep learning, machine learning, and neural networks. So what is the relationship between them?

Let's start with artificial intelligence. Artificial intelligence was proposed at a Dartmouth conference in 1956. More precisely, in 1956, scholars identified them as artificial intelligence at the meeting. In fact, some specific research on it has already begun.

So artificial intelligence is already an area with more than 60 years of history. Why has artificial intelligence gradually entered the public eye in recent years? In fact, the wave of artificial intelligence in these years has been the third heat of artificial intelligence, and far exceeds the previous heat. The first and second waves disappeared into the long river of history. They encountered their own problems in those days, such as the first time because they encountered more complicated needs after perfecting some artificial intelligence projects. At that time, it was discovered that the artificial intelligence technology at that time did not have the ability to solve it, which led to the world's main research countries interrupting the research funding of artificial intelligence. The first development was suspended because of lack of money. In essence, technology encountered a bottleneck. In the second wave, because the expert system brought great effects and benefits to many enterprises, and artificial intelligence was ready to develop vigorously, there was no space for the strong development of the fourth generation computer, and almost all of the funds And manpower has been invested in the development of the fourth generation of computers.

In addition to the influence of external factors, artificial intelligence itself has some problems. The two core problems in the early days were data and computing power, because at the time there was a very good fitting algorithm model, but it was worn over-fitting. Hats are model dimensions, but there is no detailed and sufficient data support.

In fact, after 1956, the development of artificial intelligence was very fast, and solved many problems in the industry. At that time, scientists at the time thought that at the time of development, the machine might reach the human level in about 20 years. Looking at it now, the science giants at the time were still overly optimistic. Until 60 years later, we are still in the field of weak artificial intelligence.

Deep learning, machine learning, neural networks, what is the relationship between them?

Figure 1 Development process of artificial intelligence industry

Speaking a bit far, back to today's topic, the relationship between artificial intelligence, machine learning, neural networks, and deep learning.

The concept of artificial intelligence may be a big pit, confusing many people. To explain briefly, artificial intelligence is the realization of what humans can do. This is the purpose. There are many details, the most central of which we can understand as part of the human brain, that is, machine learning.

Deep learning, machine learning, neural networks, what is the relationship between them?

Figure 2 artificial intelligence diagram

Industry researchers at Innov100 believe that machine learning can be simply understood as the core method of implementing artificial intelligence. He is not a single method, but a collection of many algorithms. That's right, the core of artificial intelligence is supported by various algorithms. However, the current machine learning is easier to understand as a simple semi-artificial intelligence algorithm. For example, when we visit a treasure, there will always be a column to recommend various products, or after you browse some products, you will find the home page. Even the default search terms have become the keywords of the products you are viewing. This is a fusion of machine learning-based recommendation algorithms, and also a portrait of the user in the background to more accurately predict the products you want to buy. In fact, there are still some problems behind the implementation of such technology, such as your privacy, if you are predicted to be accurate, then you have any privacy, all your operations may quietly betray you.

Neural Network is simply one of many algorithms for machine learning. It is designed to mimic the processing of the human brain. It is hoped that it can operate according to the logic of the human brain (although the research on human brain is still not thorough enough at present) ). Neural networks have been around for many years, but they are rarely heard now. Researchers at the Innov100 industry believe that neural networks can be easily divided into single-layer, double-layer, and multi-layer networks. There are a lot of problems in the neural network before, the number of layers can't be too much, there are too many parameters to adjust, and the sample data is too small. In short, it used to be a technology that was not optimistic. Until 2006, Hinton published a paper in Science and related journals, and for the first time proposed the concept of “deep belief network”.

DeepLearning is actually an extension of the neural network. It is proposed from the concept and gradually shows its talents in the field of artificial intelligence. Especially in 2012, it achieved amazing results in the field of image recognition. Like neural networks, deep learning is also a collection of algorithms, except that the algorithms here are based on new algorithms for multi-layer neural networks. He is a new algorithm and structure. The most famous of the new network structure is CNN. It solves the problem of too many traditional deep network parameters and is difficult to train. It uses "local receptive field" and "power planting." The concept of sharing greatly reduces the number of network parameters. The key is that this structure does fit the working principle of visual tasks on the human brain. There are many new methods: new activation functions: ReLU, new weight initialization methods (layer-by-layer initialization, XAVIER, etc.), new loss functions, new over-fitting methods (Dropout, BN, etc.). These aspects are mainly to solve some of the shortcomings of the traditional multilayer neural network: gradient disappearance, over-fitting, etc.

Because it solves some of the legacy problems of early artificial intelligence, under the blessing of big data and great computing power, artificial intelligence re-enters the public's vision. And become the core technology in visual recognition, image recognition, speech recognition, and chess AI. So now deep learning is a new neural network, the essence of which is still the neural network, but it is different from the old neural network. In addition, neural networks are rarely discussed now.

I hope that after reading the article, you can understand the relationship between artificial intelligence, machine learning, neural networks and deep learning. In addition, if you continue to expand on this basis, there will be migration learning and reinforcement learning, which will continue to be explored in later articles.

Replacement Bulb Lamp

The replacement bulb lamp is a substitute bulb for the projector. The price of the replacement bulb lamp will be more affordable than the original Projector Lamp, and the average life can reach 3000 hours, which is suitable for basic film and television in daily life. The operation is simple and easy to install.

Replacement Bulb Lamp,Projector Bulb,Replacement Original Bulb Lamp,Projector Bulb Lamp

Shenzhen Happybate Trading Co.,LTD , https://www.happybateprojector.com