The progress of human civilization is a thing step by step to liberate people from their need to do repetitive work out, do what they want to do more.
This requires businesses to achieve automation, intelligent (can be considered automatic when advanced stage).
I do not know if you have people you said, when you do the same thing on the computer more than twice, you will need to write a program to do it. This is the method used to automate procedures.
But the question arises, some tasks difficult to use the program. For example, to identify any light at any angle of view three-dimensional object in a two-dimensional picture, this task can not be achieved with either program, either written procedures will be very complicated.
Then there was machine learning, no handwritten code / program, only need to collect a large number of examples relevant to your task / data, you can get the job done these examples from the program / data by machine learning methods.
It should be noted, obtained by the machine learning programs and traditional hand-written program is very different, the resulting machine learning programs (such as the depth of the neural network model) may contain millions of parameters. If we do the right, the resulting machine learning program can respond to new input data, get the correct output. When the input data is changed, the machine learning program can be obtained by using new input data is retrieved and re-training.
Now training a lot of data to get a new machine learning models and parameters (that is, the program obtained by the machine learning), in particular on the following tasks, than hiring a specific program handwriting to complete the task, there is a distinct advantage.
Mode (pattern) recognition: recognition as the picture of the object, recognize human faces, identifying people say.
Identify anomalies: determining whether the current situation and the normal situation has obviously not the same, such as abnormal transfer transactions, nuclear power plants and other abnormalities of the sensor data.
Prediction: For example, to predict what would you buy something, then recommend to you; forecasting exchange rates and stock prices.
Overall, machine learning is a method by which method you can design a super multi-parameter model to complete the task (solve the problem), the model parameters can be obtained through training data.
Summary: We need this machine learning method, because it than people write programs to solve specific problems more efficiently and more efficient than using manpower to repetitive tasks again and again to complete the task. Machine Learning> people write programs> artificial
More introduction to machine learning, to view a video of the cattle Geoffrey E. Hinton explain https://www.cs.toronto.edu/~hinton/coursera/lecture1/lec1a.mp4
Zu Guoxiang, in Shanghai
