Source | AI algorithms and image processing
main content
What is Machine Learning
Applications of Machine Learning
Components of Machine Learning
Machine Learning and Other Fields
What is Machine Learning
What is "learning"? Human learning is through observation, the accumulation of experience, master a skill or ability. Just as we grew up learning to identify letters, recognizing Chinese characters, it is the process of learning. And Machine Learning (Machine Learning), as the name suggests, is to make a machine (computer) can be the same to humans by observing the large amount of data and training, found that the law of things, get some sort of problem analysis, problem-solving skills.
Machine learning process starting from data, calculated after the computer eventually get a certain kind of performance.
For example, through the computer so that the computer learn to predict the stock. Using the data a decade ago to learn, tell me tomorrow that how to invest, if the machine did, then that machine really learned something. It is our hope that machine learning can do.
Example: How to identify a tree
A program designed to define a tree, it is very difficult
By studying the data to identify a tree (3 years old children can do)
Machine learning system designed to identify a tree than a program easier and more
Through our brain to analyze these things more difficult, because you want to learn myself and analyzed by the machine and found that these laws.
Machine learning to build a complex system of time is an optional method:
such as:
When humans can not do a complex system, all the laws are written clearly, such as robotic exploration of Mars, no way to predict robot on Mars What will happen, for example, encountered a pit, how to do, which requires its own machine to learn how to deal with
No way to define a rule when - voice / visual recognition
Some people thought the application - stock market trading
In the large-scale data a user oriented issues - individual users market
Machine learning to do, is to teach a computer similar to fishing methods, it can be good enough for a lifetime.
So under what scenario, consider using machine learning to solve it?
Three key conditions Machine Learning:
There are some potential things law itself can go to learn, and have clear goals
There are some things the rules, but we do not know how to put it down to write the code
Machine learning materials needed, otherwise the machine does not know how to learn
When the three key conditions are met before considering the use of machine learning
Test & answers:
1, forecast the next few minutes after the child cry? no (no rules)
2, to determine whether there is a round figure? no (rules can be easily defined)
3, to determine whether the user to distribute the letter with the card? yes, is not easy programming, a large customer history information
4, when the Earth will be destroyed? no (not enough data)
Applications of Machine Learning
Machine learning in all aspects of our basic necessities have applications.
（1）Food：
Source: Twitter (Evaluation positioning +)
Function: How do you know this is unfortunately the taste of food
（2）Clothing
Data: Product images, user outfit
Function: Tell us how to match clothes can be more Fashion
（3）Housing
Data: architectural features of the house, use of energy
Features: the ability to predict the energy saving houses above
（4）Transportation
Data: traffic light images and meanings
Function: to accurately identify traffic light signal
The above are just a few examples, of course, there are many examples like this!
Application of machine learning in education
Data: Some student records online lesson, and answer records and school records
Function: to predict what the student union, what will not, and recommended some information.
Then the machine learning how to design it?
Data from 3000 are given nine million students
Using machine learning to automatically determine the difficulty of the problem, etc.
Entertainment applications, recommendation system
Data: How many users like what movie
Function: predict how much a user likes a probability not seen the movie
So the computer is to learn how these features do?
Using the model and the user to describe the movie with a bunch of features, find the inner product of two strings feature, if multiplied score high, give a very high recommendation scores. But we have no way to define these characteristics, so the machine learning by past data, to learn these characteristics, and predict the user how much they like the movie.
Test & answers
Machinery used in the following areas which can not?
1, the financial (market forecast)
2, medical (predicted efficacy)
3, legal (automatically gives a summary from public documents, easy to find and read)
Any yes 4, not above
Components of Machine Learning
How to formulate the problem of machine learning
Basic terms:
Enter: x (user behavior)
Output: y (forecast based on the results of good / bad, decide whether or not to card issuers)
The objective function: f, unknown rule ---> objective function
Data, training samples (data collected in the past)
Hypothesis, the hypothesis choose the best moment of the corresponding function is called g, g that best represents the inherent law of things, we also want to get the final model expression
Machine Learning flow chart:
Data from unknown laws by learning algorithms to dig, so close to the final g f
important point:
The objective function, f is unknown
Hypothesis is to g as close to f, but may still differ from f
For example, a credit card, for example
g in the end look like
Today decide whether or not to the customer credit card, here are some possible formulas
h1: annual income does not exceed 800,000 more than give
h2: more than 100,000 to the credit card debt
h3: job less than two years to give him the credit card
All possible h, g are put in the collection, and the most likely outcome from the collection found
Model = + learning algorithm hypothesis
Definition of machine learning: starting from data, machine learning algorithms to calculate a hypothesis (hypothesis) g, g we hope this to be very close to what we most desire that f.
Test & answers
Recommended songs
Machine Learning and Other Fields
Machine learning-related fields are:
Data Mining (Data Mining) -> dig some useful information from the data
AI (Artificial Intelligence) -> calculate and show something very clever behavior (such as AI chess)
Statistics (Statistics) -> use data to make some inferences (such as coin issue)
Machine learning and data mining, are very much alike in some places is still the same, inseparable
Machine learning is one way of implementing artificial intelligence
Statistics is one way to achieve machine learning, statistical results are given more attention to the use of mathematical theory, few concerns calculation. Statistical machine learning to provide a lot of useful tools.
Test & answers
to sum up
This lesson introduces the main concepts of machine learning, machine learning is actually a function from the data found, then the function we most desire and goal is to find very close. Machine learning has applications in many places, the core algorithms, data (information), hypothesis (hypothesis) finally got g.
Machine learning and data mining, artificial intelligence, statistical relationship between these three areas to be compared, and in various fields, they each have their own orientation.
References:
https://www.bilibili.com/video/BV1Cx411i7op?p=1
https://blog.csdn.net/red_stone1/article/details/101303228
【end】
◆精彩推荐◆
今晚8点，我们一起来看Sophon KG如何追寻新冠病毒轨迹，运用AI技术、工具建立相关知识图谱，通过确诊案例的亲属、同事和朋友的关系网找出密切接触者进行及时隔离，同时刻画出确诊案例的活动轨迹，找到其关系网之外的密切接触者及病毒可能的“行凶环境”。推荐阅读
Dual regression network, up to the latest SOTA for single image super-resolution | CVPR 2020
"Dead carry" high concurrent flow, barley grab votes technology Nirvana road
Bitcoin from the "masked men" to create that next "Bitcoin" would create a masked man?
On the continued deployment and best practice built on the container
Your point of each "look", I seriously as the AI
