Review of basic knowledge points for AI final exam (Introduction to AI)

Chapter 1 Introduction

General explanation: Artificial intelligence is intelligence realized on machines (computers) using artificial methods, or is called machine intelligence or computer intelligence.
Knowledge & Intelligence
Knowledge is the understanding of the regularities of the objective world that people know through experience, learning or association, including facts, conditions, processes, rules, relationships and laws, etc.
Intelligence is the ability to apply knowledge to deal with certain environments or problems or the ability to think abstractly.
The short-term goal is to study the use of machines to imitate and perform certain functions of the human brain, and to develop related theories and technologies.
There are three stages in the development of artificial intelligence: [Computation] stage, [Perception] stage, and [Cognition] stage.
Turing (1912-1954) first proposed that "machines can think" and is known as the "Father of Artificial Intelligence".
Dartmouth Conference : The first artificial intelligence seminar in human history in 1956 , marking the birth of the artificial intelligence discipline.
Feigenbaum develops the first expert system DENDRAL
The Chinese Society of Artificial Intelligence was established in Changsha in 1981 with great difficulty. After that, it was not recognized by the domestic scientific and technological circles for a long time. It could only be affiliated with the Institute of Philosophy of the Chinese Academy of Social Sciences. It was not until 2004 that it was able to "return to its ancestors" and be affiliated with the Institute of Science and Technology of China. Technology Association.
In 1989, the China Joint Conference on Control of Artificial Intelligence ( CJCAI ) was held for the first time .
The three major cognitive views of artificial intelligence (three schools of thought) :
① Symbolism (derived from mathematical logic, functional simulation methods, cognitive primitives are symbols)
② Connectionism (bionics, structural simulation, neurons,)
③Behaviorism Actionism (cybernetics, behavioral simulation, perception-action )
Research goals of artificial intelligence.
Short-term goal: to build intelligent computers to replace part of human intellectual labor.
Long-term goal: reveal the fundamental mechanism of human intelligence, and use intelligent machines to simulate, extend and expand human intelligence.
Research methods of artificial intelligence
① Functional simulation method, symbolism school, express the problem as a certain logical structure to realize artificial intelligence function
② Structural simulation method, connectionism school, simulate the human brain according to the physiological structure and working mechanism of the human brain Intelligence
③ behavior simulation method, behaviorism school, "perception-action" model
④ integrated simulation method, close cooperation to learn from each other's strengths and weaknesses

Chapter 2 Knowledge Representation Method★

Let’s first understand a few basic concepts through a typical example:
Knowledge representation is the basis for problem solving
Problem solving is one of the core issues of artificial intelligence
In intelligent systems, knowledge is a description of the world, which determines the system's capabilities; representation is the encoding of knowledge, which determines the performance of the system.
Attributes of knowledge:
①Authenticity: Knowledge can be proven true or false through practice or logical reasoning.
②Relativity: There is no absolute truth or falsehood in general knowledge.
③Incompleteness: It is divided into two categories: incomplete conditions and incomplete conclusions.
④Uncertainty: ambiguity and imprecision, for example: people may become confused when they are old.
⑤Representability: Knowledge exists in the human brain as human experience and can be expressed in various ways.
⑥Storability, transferability and processability (such as books, data processing)
⑦Compatibility: All knowledge existing in one body is not inconsistent with each other
Basic methods of knowledge representation : state space representation, problem reduction representation, predicate logic representation, semantic network representation, framework representation, ontology technology, and process representation.
①State space method: a problem representation and problem solving method based on solution space
②Problem specification: Starting from the goal (problem to be solved), reverse reasoning, establishing sub-problems and sub-problems of sub-problems , until finally the initial problem is reduced to a set of trivial primitive problems (problems that can be directly solved).
AND-OR graph : A structural graph composed of AND nodes and OR nodes. Its terminal leaf node corresponds to the original problem of the original problem.
Terminal nodes are solvable nodes (because they are related to the original problem), and non-terminal nodes that have no descendants are unsolvable nodes.
What are the minimum number of steps required for the N-order Tower of Hanoi? ( 2^n-1 )
③Predicate logic representation (first used to represent knowledge in artificial intelligence)
(1) What is a predicate? The components in atomic propositions that describe the properties of individuals or the relationships between individuals
(2) The basic components of predicate logic: predicate symbols P/Q/R…, function symbols f/g/h…, variable symbols x/y/z…, Constant symbols a/b/c…
(3) Use conjunctions to combine atomic predicate formulas into compound predicate formulas, which are called molecular predicate formulas [Conjunctions: ∧ (AND, AND, conjunction); ∨ (OR, or, disjunction); implication →; ~ (non, Not)]
(4) By using conjunctions ∧, ∨ and => (implication, or implication) , multiple atomic formulas can be combined to form a more complex formula . ( The atomic predicate formula is also a well-formed formula .)
(5) Predicate formula-property:

negation of negation~(~P) ≡ P
P∨Q ≡~P→Q
Morgan’s law~(P∧Q) ≡~ P∨~Q
~ (P∨Q) ≡~ P∧~Q
distributive law P∧(Q∨R) ≡ (P∧Q)∨(P∧R)
P∨(Q∧R) ≡ (P∨Q)∧(P∨R )
Commutative law P∧Q ≡ Q∧P
P∨Q ≡ Q∨P
associative law (P∧Q)∧R ≡ P∧(Q∧R)
(P∨Q)∨R ≡ P∨(Q∨R)
Inverse Law P→Q ≡ ~ Q→~P

(6) Replacement and unification
# Replacement
Pseudo-element reasoning: The operation of generating the well-formed formula W2 from the well-formed formula W1 and W1→W2.
Universal reasoning: From the well-formed formula (x)W(x), the well-formed formula W(A) is generated, where A is an arbitrary symbolic constant.
#Unity:
Find the substitution of the item on the variable so that the two expressions are consistent, which is called unification.
If there is a permutation s such that: E1s=E2s=E3s=…, then this s is called the union of {Ei}, because the function of s is to make the set {Ei} into a single form.

④Semantic network representation (a marked directed graph composed of nodes and directed arcs)
The essence of multi-semantic network representation: transforming multi-relationships into a combination of a set of binary relationships, or a binary relationship The conjunction of. ( Convert multivariate relationships into binary relationships by adding additional nodes )
★Two reasoning mechanisms of semantic networks: matching and inheritance★
Three types of inheritance:
direct pass (pass)──child nodes directly inherit the attributes of parent nodes and
additional pass ( add)──The child node combines the attributes of the parent node with its own attributes. Exclude
transfer (exclude)──The attributes of the child node and the parent node are incompatible.
Summary of suppressed transfer:
⑤ Frame representation (proposed by American artificial intelligence scholar Minsky in 1975 )
is also a semantic network, and inheritance is its important feature
. A semantic network can be regarded as a collection of nodes and arcs, or as a collection of frames. Usually the framework uses nodes, slots, and values
in the semantic network to represent the structure. A frame describes a type of object , consisting of the frame name and some slots that describe various aspects of the object. Each slot can be divided into multiple facets. Each facet can have one or more values [remember the corresponding position] slot It can also be another frame on the side, forming a framework. The connection between the network frames is indicated by the slot name. Common slots: ISA slot, Subclass slot, AKO slot, Instance slot (the inverse relationship of the AKO slot), Part-of slot, Infer slot ( Point out the logical reasoning relationship between the two frameworks, which can represent the corresponding production rules) and the Possible-Reason slot (opposite to the Infer slot). The two reasoning activities represented by the framework: matching and slot filling.
⑥Ontology technology
Ontology: formal explanation and specification of "conceptualization" .
Theoretically, ontology refers to a " formal , clear and detailed description of a shared conceptual system ." In other words, ontology is actually a formal expression of a set of concepts in a specific field and their relationships with each other. Ontology is usually used to describe domain knowledge. Ontology is not the same as an individual, it is the consensus of the group. It can be understood like this: ontology is a conceptual model abstracted from the objective world . This model contains basic terms and relationships between terms in a certain subject area (or concepts and relationships between concepts). Classification of ontology: According to the ontology application theme, ontologies are divided into five types: domain ontology, general or common sense ontology, knowledge ontology, linguistic ontology and task ontology . According to the level and domain dependence of ontology, Guarino et al. divide it into four categories: Top-level ontology : studies general concepts and relationships between concepts, such as space, time, events, behaviors, etc., which has nothing to do with specific applications and is completely Independent of a defined domain and therefore can be shared on a larger scale. Domain ontology : studies concepts and relationships between concepts in a specific domain. Task ontology : Defines some general tasks or related reasoning activities to express concepts within specific tasks and the relationships between concepts. Application ontology : used to describe some specific applications, which can refer to specific concepts in the domain ontology or concepts appearing in the task ontology.
⑦ Process representation (based on the goal of solving the problem, according to the laws of the development process of things, using relevant knowledge to design and describe the solving process) divide the overall goal
of solving the problem into process (Procedure) goals , and then combine them with the knowledge utilization link to determine It is a number of operation steps, expressed as a process. Each process is a program that is used to complete the processing of a specific event or situation.
Follow-up or to be updated…