Chapter 1 Overview of Artificial Intelligence
1. What is artificial intelligence?
Artificial Intelligence (AI), also known as Machine Intelligence (MI, Machine Intelligence),
mainly studies the use of artificial methods and technologies to develop intelligent machines or intelligent systems to imitate, extend and expand human intelligence, biological intelligence, and natural intelligence. , to realize the intelligent behavior of the machine.
Brief summary: Simulate or realize human intelligence with machines.
Note: The definition of artificial intelligence is divided into four categories:
(1) systems that think like humans
(2) systems that act like humans
(3) systems that think rationally
(4) systems that act rationally
2. What are the goals and significance of artificial intelligence research?
Short-term goal: To achieve machine intelligence partially or to some extent, making existing computers more flexible, smarter and more useful.
Long-term goal: to manufacture intelligent machines, so that they have perception and interaction abilities such as seeing, listening, speaking, and writing, advanced thinking abilities such as association, learning, reasoning, and understanding, and abilities of analysis, problem solving, and invention.
3. What are the basic technologies of artificial intelligence?
(1) Knowledge representation technology
(2) Knowledge reasoning, calculation and search technology
(3) System implementation technology
4. When, where and how was artificial intelligence born? History
In the summer of 1956, ten experts from mathematics, psychology, neurophysiology, information theory and computers held a two-month meeting at Dartmouth University in the United States to discuss issues related to machine intelligence. At the meeting, McCarthy proposed to use " The term "artificial intelligence" marks the official birth of the discipline of artificial intelligence.
Development:
(1) reasoning period
(2) knowledge period
(3) learning period
5. Three major research schools, approaches and methods of artificial intelligence
Traditional division method:
- Symbolicism (Functional Simulation)
- Connectionism (Structural Simulation)
- Evolutionism (behavior simulation)
modern division method:
1. Symbolic intelligence genre
2. Computational intelligence genre
3. Swarm intelligence genre
Chapter 2 Graph-based Knowledge Representation and Graph Search Technology
1. Knowledge required for problem solving
(1) Narrative knowledge (describes the problem state)
(2) Procedural knowledge (describes the transition between states)
(3) Control knowledge (describes how to choose the appropriate operation in the current state)
2. Commonly used knowledge representation?
State-space graph
AND-OR graph
Predicate logic
Productions
Framework
Semantic network
3. Search strategy
According to whether there is guidance, it can be divided into blind search and heuristic search.
Blind search : unguided search, also known as exhaustive search.
In the search process, without any background knowledge as a guide, without considering any information related to the solution, the nodes of the tree are randomly generated in a predetermined order (such as breadth-first and depth-first), and judge whether it is a solution, until Find a solution or prove that the problem has no solution.
(For large or infinite state space problems, blind search is too inefficient)
Heuristic search : A search process that uses "heuristic information" as a navigation. "Heuristic information" refers to the information or knowledge related to the problem that is conducive to finding the problem as soon as possible, such as the distribution law of the solution to the problem to be solved, the experience and tips for solving the problem, etc. (Game, machine learning, data mining, intelligent retrieval, etc.)
4. Advantages and disadvantages of depth-first and breadth-first search
Depth-first has high efficiency and may enter infinite branches. If the problem has a solution, it is impossible to find a solution, and it cannot be guaranteed to be the optimal solution.
Breadth priority is inefficient, but if the problem has a solution, the solution will be found, and it is the optimal solution.
5. The role of OPEN table and CLOSED table
OPEN table: dynamic data structure, register and record the nodes currently to be investigated.
CLOSED table: dynamic data structure, record the nodes that have been inspected.
Chapter 3 Graph-based Knowledge Representation and Graph Search Technology
1. Resolution principle
The resolution principle was proposed by Lu Bingxun in 1965, also known as the resolution principle,
usually using the resolution inversion method (contradictory method)
2. Find the clause set process
(1) Eliminate implication words and equivalent words
(2) Narrow down the scope of negative words, so that negative words only act on atomic formulas.
(3) Variable standardization
(4) Eliminate existential quantifiers and perform variable substitution at the same time:
A. Existential quantifiers are no longer within the scope of universal quantifiers. At this time, existential quantifiers are eliminated and a new individual constant is used;
B. Existential quantifiers appear in universal quantifiers Within the jurisdiction, the existential quantifier is eliminated at this time, and the constraint variable within the scope of the quantifier is replaced by a function of the variable constrained by the universal quantifier. This function is called the Skolem function.
(5) Eliminate all full quantifiers
(6) Formulate into conjunctive normal form
(7) Rename appropriately so that there are no guiding variables with the same name between clauses.
(8) Eliminate conjunctions and form a set S with clauses as elements.
3. Resolution strategy
(1) Simplification strategy
Deletion strategy (delete newly generated classes included in existing ones)
(2) Restriction strategy
Support set strategy (at least one clause that is not in G or descendants)
Linear resolution strategy (must have one that is Introduced)
ancestral filtering strategy (either at least one is in the original clause set, or one is another ancestor)
semantic resolution strategy
(3) ordering strategy
unit resolution strategy (must have a unit clause)
4. Advantages and disadvantages of reasoning by reasoning from principles
Advantages: Single form, very simple processing rules
Disadvantages: Many important semantic information is lost in the process of transforming clause sets, and only a logical relationship in form remains, corresponding to the heuristic search and human-computer interaction. Many difficulties.
5. Machine reasoning method based on predicate logic
Natural deductive reasoning
Inductive deductive reasoning
Rule-based deductive reasoning
6. Rule-Based Deductive Reasoning
Forward deductive reasoning
Backward deductive reasoning
Bidirectional deductive reasoning
Chapter 4 Representation and Reasoning Techniques for Uncertain Knowledge
Uncertainty reasoning:
It refers to the reasoning process that starts from uncertain initial evidence and uses uncertainty knowledge to draw a reasonable or basically reasonable conclusion while maintaining a certain degree of uncertainty.
Uncertainty reasoning = symbol deduction + uncertainty calculation
1. Uncertainty types and characteristics
(1) Random uncertainty
means knowing all the results that will happen, but not knowing which one will happen, and knowing the probability of each occurrence.
(2) Fuzzy uncertainty
means that there is no standard (for example: Xiao Wang is tall).
(3) Incompleteness
refers to incomplete understanding of something.
(4) Inconsistency
refers to the incompatibility of conclusions over time.
2. What problems should be solved in uncertainty reasoning?
(1) Representation and measurement of uncertainty
(2) Matching algorithm of uncertainty (
3) Calculation and propagation of uncertainty
Chapter 6 Machine Learning
1. Definition of learning and machine learning
Learning: The enhancement and improvement of the system's own capabilities during repeated work, so that the system will perform the same or similar tasks better or more efficiently than it does now when it performs the same task next time (Simon).
Machine Learning: A computer program that implements the behavior of improving performance on a task through experience.
2. Basic structure and function of machine learning system
Note: The model can be replaced by a knowledge base
Environment: Provide external information
Learning link: Process information provided by the environment and receive feedback from the execution environment in order to improve knowledge in the knowledge base and meet performance requirements
Knowledge base: Learned knowledge
Execution link: Test the performance of learned knowledge
Chapter 8 Natural Language Processing
1. What is Natural Language Processing?
It refers to the use of computers to analyze and process natural language, so that computers can understand and express natural language, and realize natural language communication between humans and computers. (IBM Watson Core)
2. The levels involved in natural language processing (language has two attributes of speech and text)
(1) Speech analysis:
(2) Lexical analysis: recognition and part-of-speech judgment
What methods are there for Chinese word segmentation
(1) Word-based method (matching with existing vocabulary)
(2) Character-based method (according to the mark the location in and then scan)
(3) Syntactic analysis (grammatical analysis): Judging whether it is legal
is divided into syntactic structure analysis and dependency analysis
(4) Semantic analysis: the meaning of a piece of text
A. Word meaning disambiguation
Method:
knowledge- based method Based
on supervised learning
The method of supervised learning
B. Semantic role identification
(5) Pragmatic analysis
3. Characteristics of natural language
(1) New words appear constantly, and it is difficult to fully include them in the dictionary
(2) The expression of natural language is very flexible and difficult to fully formalize
( 3) Natural language is full of ambiguities, and it is difficult to completely resolve
pronunciation, word segmentation, part of speech, syntactic structure, and word meaning ambiguity
(4) There are various language innovations in natural language, which are difficult for machines to cope with
3. Applications of Natural Language Processing
(1) Speech recognition and synthesis
(2) Machine translation
(3) Information retrieval
(4) Question answering system
(5) Information extraction
(6) Text summarization
(7) Text classification
(8) Social computing
4. Language knowledge base
(1) Grammatical Knowledge Base (Modern Chinese Grammatical Information Dictionary)
(2) Semantic Knowledge Base (Semantic Network -> HowNet)
