Don’t confuse machine learning with artificial intelligence!


Intelligence is the ability to think rationally and control behavior. Human beings have the wisdom to think and use common sense to make decisions. Artificial intelligence is a research field of building intelligent agents, so the artificial intelligence we build in the future can think and act rationally like human beings. Turing test was proposed by Alan Turing (1950) to provide a satisfactory definition of intelligent operation. The robot can pass the Turing test if it has the following functions:

  1. Communicate with people by understanding and writing natural language;

2. Knowledge representation (know how to present knowledge to users);
3. Knowledge reasoning (know how to infer answers from stored knowledge to answer human);
4. Machine learning inference mode and adapt to the new environment.

In short, AI is to study the rules and algorithms that help to build intelligent machines. AI solves a set of problems that are NP complete.
Artificial intelligence is a wide range of research fields, involving the following five important disciplines:

  1. Expert system;
  2. Neural network;
  3. Fuzzy system;
  4. Robot;
  5. Natural language processing.

Machine learning (ML)

Machine learning is a subset of artificial intelligence, which obtains some data needed by human beings through algorithm learning in data. Learning can turn people into geniuses and adapt them to the new environment. Similarly, the learning ability of the machine makes it strong enough to adapt to the new environment. The goal of any machine learning algorithm is to maximize its goal through the learning process so that it can process invisible data.
Two key learning methods (algorithms) for machine learning are:

  1. Supervised learning: external designers or tag data are helpful for machine learning.
  2. Unsupervised learning: there is no label data or external designer for machine learning.

The goal of AI is to make machines as intelligent as people.

expert system

Expert system is a system that relies on knowledge base to solve problems. Knowledge base can be expressed in different forms, such as rules, semantic network and decision tree. The expert system is composed of knowledge base and inference engine to infer or infer knowledge from the stored knowledge base. Expert systems are used where human experts are needed to solve specific problems.

knowledge base

The rule-based expert system captures the knowledge of experts in specific fields in the form of rules. These rules form the knowledge base, and then evaluate the facts through the reasoning engine to solve specific problems. Rule example:
If the sky is clear and sunny,
Then a raincoat is not needed.


Because rules are expressed in natural language, it is easy to capture the understanding of knowledge base.


Experts have different opinions on the same subject, which makes it difficult to master domain knowledge.
The maintenance and update of rules is a long process.
And there are different types of expert systems in different fields, such as rule-based expert system, fuzzy expert system and framework based expert system.


Reasoning in expert system is carried out through forward or backward link. Forward link is a data-driven reasoning technology, which starts from knowing the data and advances according to the rules. Backlinking is a goal driven reasoning, which starts from a goal and advances backward to find the data supporting the goal.

neural network

Artificial neural network (ANN) is inspired by human neural system. The system works in exactly the same way as the human brain stores and processes knowledge. A neural network very similar to human brain consists of a group of highly connected neurons or nodes. Information is stored, processed and analyzed in neurons of the network. Each node or neuron can activate other neurons in the network. The link or connection between neurons is called weight. A network can contain n neurons or nodes, which can make the network very complex. A simple neural network consists of an input and output layer.
The following are different types of neural networks:

Feedforward neural network;

Convolutional neural network (CNN);

Recurrent neural network;

Long and short term memory network (LSTM).

Artificial neural network can learn by adjusting weight weight. It is this ability of neural networks that makes them suitable for machine learning. Different kinds of learning algorithms can be used in neural networks, the most prominent one is the back propagation algorithm.

Among dozens of Alibaba cloud products with limited time discounts, click collect to start cloud practice!

Author: [direction]
Read the original text
This is the original content of yunqi community, which can not be reproduced without permission.