The future of machine learning

———————— Machine learning Experience after machine learning technology sharing

This Friday (2019.03.23), I attended a technology sharing about “machine learning”. The speaker was Mr. Liu, manager of technology Department of XX Media

The most intuitive feeling I get from this technology sharing is that machine learning is likely to be the next growth point of the Internet. Whether you are a programmer or a business manager, understanding and mastering machine learning is critical to your future personal and corporate development.

Why do you say so?

One success story is in front of us: headlines

headlinesRapid rise, beyond many information software, even beyond “NetEase news”, and “Tencent News” rival

I think there are two most important reasons:

  • Have a thick skin early onThrough the web crawler, a lot of media content is captured. Even grabbed a lot of "media content comments" from competitors.In this way, the accumulation of media content was quickly completed
  • For different types of users, personalized content recommendationRecommend the user's favorite contentIs the most important metric of the system compared to other news softwareContent quality and timelinessIt doesn’t matter anymore

Perhaps Toutiao was not set up to be a news app; Instead, it is “a piece of software that gives users the content they love and want to immerse themselves in for 24 hours.” My father’s daily immersion in his recommended content about the “astronomical universe” made me want to confiscate my phone.

I think personalized content recommendation for different types of users is the core technical driver of its success. I think the more professional way to say it is:

Create a machine learning model based on user characteristics, habits, hobbies, etc. Specific content recommendations are made according to different users’ regions, usage habits and gender.

Of course, after the completion of user accumulation, Toutiao is also actively producing its own content

Poached a lot of editors from other Internet companies (but think for a moment, even if there were a lot of editors, how much content would be produced per day? A lot of content should still rely on the web crawler to grab it. But it was a success, and there should have been co-payment for grabbing content, otherwise Toutiao wouldn’t be alive today.)

Consider Tencent news, NetEase news how to compete?

  • If we follow the headlines, we will lose the necessity of our existence as a news software.
  • If we don’t do this, and let the headlines grow, we will lose our leading position in the industry, and the income reduction is secondary.If the change continues to find no direction, it may even disappear in the competition;

As for the question I raised, I think Tencent News and NetEase News, as pure news and information software, are necessary to exist:

  • To ensure the timeliness of content (after all, people who read news want to know the latest events and reports at home and abroad)

Can only think of this one, after all, is an ordinary programmer, these problems or let the corporate leaders to think about it. Do whatever the bosses tell you to do, and make money for your family. That’s what I, the miserable programmer, do…

Back to machine learning

Back to the topic of machine learning, the headline is also meant to illustrate the importance of machine learning. And then I repeat what I think:

Machine learning is likely to be the next growth area of the Internet. Whether you are a programmer or a business manager, understanding and mastering machine learning is critical to your future personal and corporate development.

Statement:

All opinions about the headlines are purely imaginary and have nothing to do with any other person or organization. Thank you.

And, by the way, is it risky for me to talk nonsense like this? But what I said was what I thought, and I wanted to write it down, otherwise I might forget it someday.

Second, technology harvest

How does machine learning work?

  • Define the target
  • Continuous optimization

By training large amounts of data and combining probability calculation, continuous optimization will eventually improve the accuracy of machine prediction

Such as:

  • Predicting that a picture is an apple can improve the probability of success of image recognition by training a large number of feature data.
  • If a red light is about to change and a cyclist is about to pass, should he speed up or stop? This requires an impact probability calculation
  • Finally, continuous optimization, improve the accuracy of prediction

A famous mathematical formula

The nice thing about it is that it’s somewhere between zero and one, getting closer and closer to zero and one

** Prediction a picture of apple: through the training of a large number of feature data, the prediction accuracy is infinitely close to 1 **

A ordering App recommendation system

The picture shows a food ordering App recommendation system for flavor recommendation by region

  • The first layer is the concept of the state
  • The second level is the concept of provinces, such as Sichuan, Shandong and Hebei
  • The third layer is the concept of cities, such as Hengshui, Tangshan, Shijiazhuang and so on

Such as:

  • When hengshui people open the App, they see the recommended menu generated after continuous training of hengshui people using data
  • In addition, if a person from Tangshan sees the menu recommended by the App and has no appetite, the system will also perform the model backtracking function to enter hebei province and generate the recommended menu generated by the whole people in Hebei province using data training

Of course, this model could still be modified.

Create a machine learning model based on user characteristics, habits, hobbies, etc. Specific content recommendations are made according to different users’ regions, usage habits and gender.

Think of paying for knowledge apps

Ximalaya and NetEase Open Courses can also increase DAU and repurchase rate in this way

  • Based on users’ preferences, we constantly recommend free content to users to enhance user engagement
  • In the process of recommending free content, occasionally recommend its accustomed paid content to increase the rate of repeat purchase

Third, programmer direction

Actually, at this point, MY article is over. However, when I communicated with zhu Zhu, a great colleague and good friend, I felt that his opinion pointed out the direction for the growth of future programmers. Here I stuck it word for word:

I feel that recommendation system is an application of machine learning, which still needs big data to train and extract features. Big data can be a direction [wry smile], and the development of big data is inseparable from distributed computing, database support. Machine learning is a small part of the artificial intelligence, and speech recognition, robot, direction of the development of computer graphics and natural language processing a lot of unnecessary focusing only on machine learning At present, we can make of natural language processing and graphics, feeling can make robots, have difficulty 😄 speech recognition, natural language processing is good, A lot of my colleagues at my first company did itCopy the code

Big data, distributed computing, databases, natural language processing, speech recognition, computer graphics, robotics — these are all possible directions for programmers. Choose a direction, deep cultivation, enough to study for a lifetime. (Choose a direction, continue to study, there will be no big success there will be small success)

Confused programmers should stop being confused

Choose a favorite direction, always maintain a positive learning attitude, continue to improve the level of technology, is the future of the programmer based on the fundamental…

Imagine the future

I think it is very possible to enter such an era

  • Solar power, machine work (electricity converted to other energy materials)
  • People don’t need to work, go into communism, according to need, just need to enjoy the sun, air, enough

Ha, smell freedom…

## finally

  • Thanks to these big factories of the Internet, we can provide a platform to communicate with these big cattle
  • Thank colleagues and friends for their opinions

========== THE END ==========

It took nearly four hours to write and revise this article. It was not easy to write the article until two o ‘clock last night