This course provides a general introduction to the basic methods and main applications of AI, including:
Basic methods of problem solving: such as heuristic search and logical reasoning
Knowledge Representation: Production Representation and Structured Representation
Introduction to Machine Learning
One of the main purposes of artificial intelligence is to solve non-trivial problems. Commonly used technologies are search technology and logical reasoning technology.
In the search process, the introduction of heuristic knowledge reduces the blindness of the search and improves the efficiency.
The typical heuristic search algorithm is Algorithm A, which introduces the evaluation function f(n):
f(n) = g(n) + h(n)
f(n) represents the minimum cost from the initial state node s to the target state node through node n.
g(n) is known, and h(n) is unknown, it needs to be estimated using a heuristic function.
Example: The 8-Puzzle Problem
Given an initial state and a target state, find the minimum movement steps required to change from the initial state to the target state.
Predicate logic and its reasoning method are one of the basic methods of knowledge representation and machine reasoning. Symbolic reasoning is the main method for solving problems based on knowledge.
An important way of symbolic reasoning is deductive reasoning.
There are two ways of deductive reasoning:
There are two ways of representing knowledge:
The production system is the most widely used knowledge representation method. In this system, production rules are used to express knowledge.
Production rules are usually used to express the causal relationship between things.
P => Q
or:
if P then Q
The rule base is a collection of production rules. The rule base is the knowledge base of the production system.
Mainly refers to the semantic network representation.
The semantic network can be expressed as a directed graph describing the relationship between things.
Learning is the main goal of AI and the basic means of acquiring knowledge. Machine learning is the self-improvement of system performance.
Dietterich proposed a simple machine learning model:
The learning component uses the information provided by the environment to modify the knowledge base. The execution component completes the task according to the knowledge base and feeds back the obtained information to the learning component.
原文：https://www.cnblogs.com/danielwong2021/p/15229229.html