Open the door to the world of AI - from artificial intelligence to computer vision
Table of contents
1. About artificial intelligence
1.1 Introduction to artificial intelligence
The first thing we need to know is the basic concept of artificial intelligence. What is real artificial intelligence? know-how technology. Artificial intelligence is one of the most popular fields in today's computer world. It involves many fields and covers the abilities of perception, learning, reasoning and decision-making.
From the perspective of practical application: the core ability of artificial intelligence is to make judgments and predictions based on given input
When was artificial intelligence AI born? What are the stages of its development so far? These are the things we need to pay attention to. Let us make a series of timelines. The 1955 Dartmouth Conference marked the birth of AI. This year is known as the first year of AI. The gate of the AI world has finally begun to be formally explored by humans. It It will accompany human civilization through a long, long history. In the process of AI development, there have been three golden periods, and each golden period is due to a new concept or algorithm being proposed
- The first peak and trough of AI: In 1957, Rosenblatt invented the first neural network Perceptron, thus AI entered the first peak period. In 1970, due to breakthroughs in computing power, it failed to complete large-scale Data training and complex tasks, thus AI entered the first trough
- The second peak and trough of AI: the Hopfield neural network was proposed in 1982 , and the BP algorithm appeared four years later in 1986, making the training of large-scale neural networks possible. These two historic breakthroughs pushed AI to the The second climax was in 1990. The AI computer DARPA failed to achieve the expected results and failed to get more financial support, so the just-emerging AI wave entered a trough again.
- The third peak of AI until today: In 2006, **Hinton proposed the "deep learning"** neural network, which made a key breakthrough in the performance of artificial intelligence. In 2013, the deep learning algorithm made breakthroughs in speech and visual recognition. This AI has entered a new era - the era of perceptual intelligence
The cornerstone of artificial intelligence development - Turing test
Turing test is a very important concept in the field of artificial intelligence. The process of Turing test includes three parties: the tester, the testee and the machine, and the separation between the tester and the testee (a person and a machine) Next, through some devices (such as keyboards), the testees are asked random questions. After many tests, if the machine allows each participant to make more than 30% misjudgments on average, then this machine has passed the test and has been awarded It is considered to have human intelligence, but this is not absolute. It does not mean that as long as a machine passes the Turing test, it is artificial intelligence. The Turing test is just a very, very important reference and standard. Needs to be judged on many fronts
1.2 Three Core Elements of Artificial Intelligence
The three core elements of artificial intelligence: data, algorithm, computing power
data
One of the elements of artificial intelligence - data, refers to the data we input into the machine, and the machine makes corresponding judgments and predictions based on our input data. With the development of the Internet of Things, information and data about the entire world are becoming more and more The more, this provides enough rich and continuous nutrition for artificial intelligence, so that the machine has enough data to learn, and then feed back the results we want
The core of big data technology is to utilize the value of data, and machine learning is the key technology to utilize the value of data
algorithm
After we provide data to the machine, the machine just got the data, and the more important thing is the algorithm. The concept of the algorithm is very simple. When we give the computer a task, we not only need to tell it what to do, but also tell it how to do it. do, and an algorithm is a series of instructions on "how to do it"
computing power
Computing power is some hardware facilities that help computers run fast and process images quickly, such as CPU, GPU and NPU (deep learning accelerator) and other hardware facilities. These hardware facilities are the key to the realization of algorithms, making the realization of algorithms fast and accurate
Breakthrough in computing power - traditional CPU and emerging computing acceleration technology + smart chip
Regarding the relationship between CPU and GPU, we can simply understand it as: CPU + parallel computing = GPU, and GPU cannot be modified after it is built. At this time, FPGA appeared to solve our hardware design problems. After completion, users can modify it at any time, which greatly reduces the risk of development
There is another very important concept that we need to understand: distributed computing, if each computer is a module, all computers assign tasks according to certain rules and then perform separate calculations, and finally combine the calculation results, this is distributed computing. operation
(The picture comes from Badou Artificial Intelligence——Wang Xiaotian of Badou Academy)
1.3 Artificial intelligence relationship circle
In the development of artificial intelligence, the rise of machine learning and the breakthrough of deep learning are milestones. What is the relationship between them and artificial intelligence? The artificial intelligence relationship circle is the most important navigator who leads us to understand the AI world
Machine Learning - A Way to Realize Artificial Intelligence
Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its application covers all fields of artificial intelligence. It mainly uses induction and synthesis rather than deduction
Deep Learning - A Technique for Implementing Machine Learning
Deep learning is the use of deep neural networks to process modules more complexly, so that the model can understand data more deeply. It is a method based on representational learning of data in machine learning. Its motivation is to establish and simulate the human brain for analysis and learning. A neural network that mimics the mechanisms of the human brain to interpret data such as images, sounds, and text
The essence of deep learning is to learn more useful features by building a machine learning model with many hidden layers and massive training data, so as to ultimately improve the accuracy of classification or prediction
Artificial Neural Networks - An Algorithm for Machine Learning
A neural network generally has an input layer->hidden layer->output layer. Generally speaking, a neural network with more than two hidden layers is called a deep neural network. Deep learning is a machine that uses a deep architecture like a deep neural network. study method
Machine learning is to achieve artificial intelligence, deep learning is a kind of machine learning
Classic Machine Learning Process
First of all, we need to prepare the training data (including a lot of category labels, that is, some feature sets that can best represent such things), and then we need to design the learning algorithm according to our desired needs, and use the learning algorithm to train the data. It becomes one of our models (such as decision tree, neural network, etc.), after we pass new data samples (prediction data) into the model, the model will give us feedback data based on its learned ability
Understanding Neural Networks (Thinking Humans - Biological Neural Networks)
What the human neural network can accomplish can be roughly explained as: After receiving an external stimulus , the neuron will convert the stimulus into information and transmit it to other functional cells, and these cells will make corresponding outputs after receiving the information
The artificial neural network is a network structure established by simulating the biological neural network. Each neuron in it must first have input data (stimuli), and a sensor that receives and processes these data , and finally makes an output . These neurons are mixed together to form our artificial neural network. The logical structure of the artificial neural network includes: input layer, hidden layer and output layer. The function of each layer is the process we mentioned above. These processes After cooperating and communicating with each other, we can achieve the desired effect
But there is more than one hidden layer. When an artificial neural network has multiple hidden layers, it is called a deep neural network . Machine learning based on a deep neural network is deep learning.
2. Computer Vision
2.1 Getting to know CV for the first time
Wan Jian returns to the sect! Before learning computer vision, we must first understand what computer vision is. To sum it up in one sentence, computer vision is to allow computers to have the ability to see, understand, and think. It can be called Computers have vision, computer vision! (It should be noted that not only the computer must have the ability to "see", but also the ability to "recognize" and "think")
2.2 Deep learning and CV and the application of CV
Computer vision based on deep learning is to enable computer vision to have the ability of "recognition" to achieve the purpose of artificial intelligence, so how to make it have this ability? Obviously, machine learning is used to enable machines to recognize images. CV is a concept that intersects with AI, ML and DL. We will use machine learning and deep learning to realize the function of our CV.
CV has 5 major applications:
1. Image classification (using convolutional neural network CNN) - linear rectification layer RELU (function f=max(0,x)) and pooling layer POOL (according to the input parameters to obtain the maximum value of each part)
2. Target detection (using R-CNN) - extract the region of interest and then perform ConvNet (convolution)
3. Semantic segmentation (using FCN network) - segment objects of different categories and then color them (for approximate categories)
4. Instance segmentation - Segment and color the objects of different instances (for instance objects)
Copyright statement: The above learning content and pictures come from or refer to - Badou Artificial Intelligence
If the article is helpful to you, remember to support it with one click~