酷兔英语
    当前位置:
  1. 首页
  2. 文章
  3. 英语能力
  4. 英语词汇
  5. 分类词汇
  6. 正文

人工智能词典(The Artificial Intelligence Dictionary) 4

0 收藏 0 分享 100 赞

章节正文

graph search algorithms
See breadth first search, depth first search, and best first search. There are many other search concepts - see the textbook by Bratko or elsewhere: look for minimax search, and alpha-beta pruning for a start, if interested.

 

H

 

heuristic
A heuristic is a fancy name for a "rule of thumb" - a rule or approach that doesn't always work or doesn't always produce completely optimal results, but which goes some way towards solving a particularly difficult problem for which no optimal or perfect solution is available.

I

 

inference engine
A rule-based system requires some kind of program to manipulate the rules - for example to decide which ones are ready to fire. (i.e. which ones have conditions that match the contents of working memory). The program that does this is called an inference engine, because in many rule-based systems, the task of the system is to infer something, e.g. a diagnosis, from the data using the rules. See also match-resolve-act cycle.
inheritance
Inheritance is a property of semantic networks and frames. It works as follows.

In the case of a semantic network, if I try to use the network to retrieve, say, the number of legs of a node (with name "Dumbo") the system will first look to see if the "Dumbo" node has an explicit "legs" link. If so, it is followed. If not, inheritance is applied, and, since "Dumbo" is an object rather than a type, the isa link is followed. Assuming "Dumbo" isaelephant, we then check the "elephant" node to see if it has a link labelled "legs". If so, we use it. If not, we look to see if "elephant" has a ako link, perhaps to "mammal". If mammal has a link labelled "legs" then we use it. If not, then we look for further "ako" links from either "elephant" or "mammal", and so on, until we either find the "legs" or run out of semantic network to search.

In the case of a frame, the effect is the same, but the details are different - an "elephant" frame would have been constructed using the information in a "mammal" frame, since elephant ako mammal. The "Dumbo" instance frame will have been constructed using the "elephant" generic frame since Dumbo isa elephant. Thus elephant inherits all the properties of mammal and Dumbo inherits all the properties of elephant. At each stage, there is an opportunity to change e.g. the number of legs if Dumbo should happen to an aberrant elephant in the matter of legs.

initial state
See article on operators and states.
instance frame
See generic frames.
isa
"isa" signifies "is a". It acts as a relation between an object and a type, and specifies that the object is a member of the type. For example, Fido isa dog signifies that the object Fido is a member of the set of all dogs.

"isa" acts like an operator in the iProlog frame implementation. For example, X isa dog with ... would have the effect of binding X to a new instance frame with all the slots of the dog generic frame, together with whatever extra information (such as slot values) was provided after the with.

"isa" should be contrasted with ako.

 

J

 

K

 

L

 

M

 

match-resolve-act cycle
The match-resolve-act cycle is the algorithm performed by a forward-chaining inference engine. It can be expressed as follows:
loop    1. match all condition parts of condition-action rules
       against working memory and collect all the rules that match;
    2. if more than one match, resolve which to use;
    3. perform the action for the chosen ruleuntil action is STOP or no conditions match
Step 2 is called conflict resolution. There are a number of conflictresolution strategies.

 

N

 

node
A component of a graph.
May also be a component of a neural network (see node in a neural network.

 

O

 

operators and states
In problem solving, the term operator is used to describe one of the procedures that can be used to move from one state to another. One starts at an initial state and uses the allowable operators to move towards the or a goal state.

The sequence of states and operators (or sometimes just the states) that lead from the initial state to the goal state is referred to as a path. The trick lies in choosing operators that do in fact lie along a path, and preferably a short or cheap path towards the or a goal state. Sometimes it is possible to navigate intelligently through state space, but sometimes blind backtracking search through state space is the only possibility.

In most cases, it is necessary to check, at each state, which operators are feasible at this particular point. If, for example, the states were physical positions in some real or simulated terrain, the operators moved one in different physical directions, it would be necessary to check for example that there were no barriers in some directions, or that there were no bad consequences for steps in certain directions. For example, a step to the West might land lead to a state which could not be escaped from with any of the available operators.

 

P

 

path
See article on operators and states.
procedural attachment
Procedural attachment refers to the practice of attaching small procedures called demons to slots in frames. Demons in frames resemblemethods in object-oriented languages such as Java, and historically precede them.

 

productionorproduction rule
See condition-action rule.

 

Q

 

R

 

ripple-down rules
Not covered in COMP9414. Discussed in COMP9416.
rule-based system
A rule-based system is one based on condition-action rules. See also backward chaining, forward chaining, inference engine, working memory, and match-resolve-act cycle.

 

S

 

search
Search is a prevalent metaphor in artificial intelligence. Many types of problems that do not immediately present themselves as requiring search can be transformed into search problems. An example is problem solving, which can be viewed in many cases as search a state space, using operators to move from one state to the next.

Particular kinds of search are described under the headings breadth-first search, depth-first search, and best-first search

semantic network
Semantic networks are a knowledge representation technique. More specifically, it is a way of recording all the relevant relationships between members of set of objects and types. "Object" means an individual (a particular person, or other particular animal or object, such as a particular cat, tree, chair, brick, etc.). "Type" means a set of related objects - the set of all persons, cats, trees, chairs, bricks, mammals, plants, furniture, etc. Possible relationships include the special set-theoretic relationships isa (set membership) and ako(the subset relation), and also general relationships like likes, child-of. Technically a semantic network is a node- and edge-labelled directed graph, and they are frequently depicted that way. Here is a pair of labelled nodes and a single labelled edge (relationship) between them (there could be more than one relationship between a single pair):

Here is a larger fragment of a semantic net, showing 4 labelled nodes (Fifi, cat, mammal, milk) and three labelled edges (isa, ako, likes) between them.

slot
A slot in a frame is like a field in a record or struct in languages like Pascal, Modula-2 and C. However, slots can be added dynamically to frames, and slots contain substructure, called facets. The facets would normally include a value, perhaps a default, quite likely some demons, and possibly some flags like the iProlog frame system's cache and multi_valued facets.

 

state
See the article on operators and states.

 

T

 

tree
see graph

 

U

 

V

 

vertex
A component of a graph. The irregular plural of vertex is vertices.

 

W

 

working memory
The working memory of a rule-based system is a store of information used by the system to decide which of the condition-action rules is able to be fired. The contents of the working memory when the system was started up would normally include the input data - e.g. the patient's symptoms and signs in the case of a medical diagnosis system. Subsequently, the working memory might be used to store intermediate conclusions and any other information inferred by the system from the data (using the condition-action rules.

The term "working memory" is also used in cognitive psychology, where it refers to the limited store of "chunks" (roughly, items in memory) available at the same time for conscious processing. 



文章标签:词典  

章节正文
Copyright © 2023 Qeto All rights reserved.
商务合作 | 联系我们 | 关于我们 | 版权声明 | 隐私条款 | 如果智培

京ICP备05059198号-1 京公网安备 11011202003401号