At the critical moment, it will be delivered to the door at the first time!
Author Zhao Ning
Making GitChat
First, the development status of artificial intelligence
1. 1 concept
According to Wikipedia, artificial intelligence is intelligence displayed by machines. Contrary to the natural intelligence of humans and other animals, AI research is defined as "agent software program" in computer science: any device that can perceive the surrounding environment and maximize its chances of success.
1.2 major event
20 16 In March, AlphaGo fought Li Shishi, who was ranked fourth in the world at that time and was a professional nine-stage player, and won with a total score of 4: 1.
On 20 16- 10, the white house released two heavy reports, namely, the preparation of artificial intelligence in the future and the strategic plan of national artificial intelligence research and development, which expounded the future development plan of artificial intelligence in the United States and the challenges and opportunities that artificial intelligence brought to government work.
VentureBeat summarized the two reports and drew seven easy-to-understand points:
1. Artificial intelligence should be used to benefit mankind;
2. The government should embrace artificial intelligence;
3. To control automatic cars and drones;
4. Let all children keep up with the development of science and technology;
5. Use artificial intelligence to supplement rather than replace human workers;
6. Eliminate the deviation in data or do not use biased data;
7. Consider security and global impact.
20 16 double eleven, luban served the double eleven for the first time, produced 65438+700 million chapters of commodity display advertisements, and increased the click-through rate of commodities 100%. If it is all done by designers, each picture will take 20 minutes, and it will take 100 designers to do it for 300 years.
20 17, the design level of luban has been significantly improved. At present, he has learned millions of creative contents from designers and evolved hundreds of millions of design capabilities. In addition, Luban has achieved the ability to make 40 million posters a day, and none of them are exactly the same.
20 17 in may, AlphaGo Master defeated world champion Ke Jie.
On June 201710/8, the DeepMind team released the strongest version of AlphaGo, codenamed AlphaGo Zero.
2017101On October 25th, at the future investment planning conference held in Saudi Arabia, Saudi Arabia awarded Sophia citizenship to the robot produced by American Hanson Robotics Company.
As the first robot in the world to obtain citizenship, Sophia said on the same day that "she" hopes to "help mankind live a better life" with artificial intelligence, and at the same time said to Musk who supports the "AI threat theory" that "people don't commit me, I don't commit crimes"!
After the meeting, Musk said on Twitter: "What is worse than inputting the movie Godfather into the artificial intelligence system?" Godfather is a classic Hollywood movie, with a plot full of betrayal and murder.
People also have to consider the ethical issues caused by Sophia being granted citizenship.
There are too many big news in the field of artificial intelligence in recent years, so I won't list them here.
Second, what are the relationships among artificial intelligence, deep learning, machine learning and reinforcement learning?
As shown in the figure, artificial intelligence is a big category, including expert system, knowledge representation, machine learning and so on, among which machine learning is the hottest and best branch at present, and machine learning also includes supervised learning, unsupervised learning, deep learning, reinforcement learning and so on.
Supervised learning is usually called classification, and an optimal model is obtained by training the existing training samples (that is, known data and their corresponding outputs) (this model belongs to a set of functions, and optimal means optimal under certain evaluation criteria).
Then use this model to map all the inputs to the corresponding outputs, make a simple judgment on the outputs, achieve the purpose of classification, and also have the ability to classify unknown data.
For example, when we were in kindergarten, we often did an activity called reading pictures. As shown in the above picture, the teacher will show us many pictures with words on them. After a long time, we will form abstract concepts in our minds, two horns, a short tail and chubby (five senses)?
Such an animal is a cow; Round, yellow, glowing, hanging in the sky? It is the sun; People look like this. When we see something similar again, we can recognize it, even if it is not exactly the same as what we saw before, but it conforms to the concept formed in our brains, as shown in the figure below.
Unsupervised learning is another learning method that has been studied a lot. The difference between unsupervised learning and supervised learning is that we don't have any training samples in advance, but we need to model the data directly.
For example, as shown in the figure, without any hints (no training set), you need to divide the following six figures into two categories. How would you distribute them? Of course, the first row is a category and the second row is a category, because the shape of the first row is closer and the shape of the second row is closer.
Unsupervised learning is to find features in data without knowing the classification of data sets.
Deep learning is a new field based on machine learning. It is a series of new algorithms, which originated from the neural network algorithm inspired by the human brain structure, and developed with the increase of the depth of the model structure, accompanied by the improvement of big data and computing power.
The concept of deep learning was put forward by the famous scientist Geoffrey Hinton and others in the articles published in Science in 2006 and 2007.
As an extension of machine learning, deep learning is applied to image processing, computer vision, natural language processing and speech recognition.
Since 2006, the research and application of deep learning in the cooperation between academia and industry has made a breakthrough in the above fields. Taking the object recognition competition in classic images based on ImageNet as an example, it beats all traditional algorithms and achieves unprecedented accuracy.
Reinforcement learning is also an important branch of machine learning, which is to learn how to do actions through observation. Every action will have an impact on the environment, and the learning object will make a judgment based on the feedback from the observed surrounding environment.
Third, how important is the mathematical foundation?
Basic knowledge of mathematics, you need high school mathematics knowledge plus high numbers, linear algebra, statistics, probability theory. Even if you don't have a perfect grasp, at least you should know the concept and know where to look when you use it.
If the foundation is not good, you can read Wu Jun's Beauty of Mathematics first, which is easier to understand. You can also learn by doing. Practice is the only criterion for testing truth. After all, most people still focus on engineering practice. If you want to be a theoretical scientist, this article is not for you.
Fourthly, the entry-level machine learning algorithm
4. 1 decision tree
Decision tree is a tree structure similar to the flowchart: each internal node represents the test of an attribute, each branch represents an attribute output, and each leaf node represents a class or class distribution. The top level of the tree is the root node.
Example: A set of data shows some people's age, income, whether they are students, credit, and whether they can buy a computer. There are three ages: young, middle-aged and old. High, medium and low income; Credit is average and good. The data and are saved in AllElectronics.csv.
Now there is a new person (data), it is necessary to judge whether this person will buy a computer.
4.2 Recent sampling
Recent sampling is to classify the existing data into several categories, calculate the distance between the newly input data and the known data, and classify the new data into which category, such as the movie classification shown in the figure below. According to the number of fights and kisses, movies with unknown movie types in the last line are closer to the romantic type and should be classified as romantic movies.
For example, irisdata.txt is a data set of Iris plants downloaded from the Internet. Classify the new data according to the data set.
4.3 Support Vector Machine
Support vector machine (SVM) is developed from the optimal classification surface under the condition of linear separability. The optimal classification surface requires that the classification line can correctly distinguish the two categories (the training error rate is 0) and maximize the classification interval.
SVM considers finding a hyperplane that meets the classification requirements, so that the points in the training set are as far away from the classification surface as possible, that is, finding a classification surface to maximize the margins on both sides.
The training samples of H 1 and H2 on the hyperplane are called support vectors, and the hyperplane is the point closest to the classification plane and parallel to the optimal classification plane.
Example: using sklearn library to realize svm algorithm, commonly known as adjusting the library, actually adjusting the library is a very simple process, and you don't even need to know the principle at the initial stage.
# code: utf-8from sklearn import svmX =,] y = [0,0, 1]clf = svm. All parameters of SVC(kernel ='linear')clf.fit(X, Y)# support vector machine can pass. Fit the function and save it in clf. Print CLF # to get support vector print CLF. Support vector # gets the CLF index of support vector sprint. Support # Get the number of support vectors in each category and print CLF. NSupport # predictdata, parameter is two-dimensional array printing CLF. Forecast (,[10, 10]]).
Verb (abbreviation of verb) book list recommendation
Wu Jun's The Beauty of Mathematics
"Machine Learning" Zhou Zhihua
Random talk on artificial intelligence information collection club
Machine learning in action Peter Harrington
TensorFlow technical analysis and actual combat in Li Jiaxuan
"Statistical learning method" Li Hang
Sixth, the misunderstanding of learning artificial intelligence-is artificial intelligence another bubble?
Artificial intelligence has been greatly exaggerated by some technology giants. In order to get the capital money, this is also reasonable, but the general public must have their own ability to distinguish and objectively analyze whether they are suitable for this industry.
Throughout the development history of the Internet, this development trend of artificial intelligence is not the first case. For example, online to offline, which was popular in 20 14, didn't know some O2O at that time and didn't dare to say that they were people in the internet circle.
Up to now, batch after batch of entrepreneurial armies have fallen, and of course giants such as Amazon and Alibaba will be left behind. Every industry has its pyramid.
In my sophomore year, it can be said that 3D printing and VR technology have been at the forefront, and various 3D printing startups and VR startups have emerged one after another. In the third year of high school, I began to fall down one after another, including I also worked on 3D printing projects. In fact, what I did was not to improve something untimely. The core framework has long been designed by the big cows.
Blindly following the trend of science and technology can only follow the trend forever.
Recently, CCTV has an artificial intelligence variety show hosted by Sa Beining, which shows that artificial intelligence has already become a red sea, and there is no essential difference with the current mobile Internet technology.
Since Google opened TensorFlow framework (there are many other excellent frameworks), many machine learning codes are tuning parameters, and some don't even need to know the principle. Of course, Daniel must have it. Again, every industry has its pyramid, but the path to the top is different.
In my opinion, there is no essential difference between calling TensorFlow's artificial intelligence development framework and calling Android's API development App. The truly great company is Google, and the latecomers are just followers.
Off-topic, I don't know if you have heard that 2 1 century is the century of biology. When this concept arose, many college entrance examination students chose biology-related majors. There was a survey on the employment direction of biology graduates from a famous university in China before, and one of the conclusions was that the best way out for biology students was to leave this major.
Of course, we have to say that biotechnology is closely related to everyone's life, but how can one afford its long development cycle? How to coordinate personal identity and social identity, self-worth and social value is also a problem that we need to think about.
Is artificial intelligence a bubble? How long will this concept last?
The sixth part is purely personal and is for reference only.
About the author: Zhao Ning, winner of the National Prize of Smart Car Competition, is the team leader. Indiegogo raised more than 1 10,000 in two months. Master's degree in automatic control (interdisciplinary and intercollegiate research) in a double-class university. Machine learning researcher, product manager, internet expert.