Data-driven, functional expert system shell

• US 4,866,634 A
• Filed: 08/10/1987
• Issued: 09/12/1989
• Est. Priority Date: 08/10/1987
• Status: Expired due to Term

• Current Assignee: Moody's Risk Management Services, Inc. (Moody's Corporation)
• Original Assignee: Syntelligence
• Inventors: Reboh, Rene, Risch, Tore J. M.
• Primary Examiner(s): Pellinen, A. D.
• Assistant Examiner(s): Williams, H. L.
• Application Number:
  US07/084,252
• Time in Patent Office: 764 Days
• Field of Search: 364/513, 364/408, 364/900
• US Class Current: 706/60
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06Q 40/03 Credit; Loans; Processing thereof
• CPC Class Codes: Y10S 706/925 Business

Abstract

An expert system shell efficiently computes functions of variables in response to numeric or symbolic data values input by a user. The system comprises a Knowledge Base in the form of a network of functions, an Inference Engine for efficiently updating values in the knowledge base in response to changes in entered data, and a Forms System that manages interaction with the user. A knowledge engineer creates the network of functions, and defines the user screens and the connection between screen objects and variables in the function network. The system allows many different types of variables, including numeric and symbolic types. The system associates a probability distribution with every variable, and computes the probability distributions for the dependent variables from the probability distributions for the independent variables. A variable can store multiple values as tables of probability distributions keyed by one or more key variables. When a user action changes the probability distributions for any variable, the system automatically maintains the specified functional relationships among all the related variables.

Claims

1. An expert system shell comprising:

(a) a computer having a display device, an entry device, and a memory for string a knowledge base;

said knowledge base comprising;

(1) variables having values represented by tables of probabitlity distributions keyed by zero, one or more formal parameters;

(2) functions defining relationships between the values of each dependent variable and the values of its corresponding argument variables;

(3) means for computing the probability distribution of the values of a dependent variable from the probability distribution of the values of the corresponding argument variables; and

(b) means for propagating the consequences of a change in the value of a variable to maintain the functional relationships among a selected subset of the dependent variables.

3. The system of claim 1 further comprising means for recording the types of variables in the knowledge base, including both numeric and symbolic types, means for automatically generating menus for variables selected with a pointing device, means for performing validity checks on variables entered via a keyboard, and means for converting formats into standard formats.

4. The system of claim 1 further comprising means for automatically generating menus for allowing the end user to see all the existing instances of a variable at runtime and means for allowing the end user to see subsets of the full menu by using a wildcard matching procedure.

5. The system of claim 1 further comprising means for identifying a variable in the knowledge base uniquely by name, said identifying means including means for segmenting the knowledge base into modules, so that variables in different modules can have the same name, thereby simplifying the creation of a knowledge base by different knowledge engineers.

6. The system of claim 1 further comprising means for displaying the uncertainty in and the expected value of any variable.

7. The system of claim 1 wherein said functions comprise a system-provided family of primitive functions, functions specifying the dependence of an instance of a parameterized variable upon other predecessor instances of that same variable, as well as upon other variables.

8. The system of claim 1 said functions comprising a coherent system-provided family of functions, including functions for (a) arithmetic operations, (b) logical operations, (c) string operations, (d) selection operations, (e) value translation or combination operations, (f) external relation operations, and (g) variable state determination.

9. The system of claim 1, said propagating means further comprising means for minimizing the computation required during propagation by assuring that the final distributions for all arguments to a function are known before the distribution for the dependent variable is computed.

10. The system of claim 1 further comprising means for assuring that the visible values or distributions of values computed for any variable depend only on the values or distributions of values of the input variables, and are completely independent of the time sequence in which the inputs are entered.

11. The system of claim 1 further comprising means for saving all data entered by the end user on a computer file, means for loading that data, and means for ensuring that the system will produce the same outputs regardless of the sequence in which the data had originally been entered.

12. The system of claim 1 further comprising means for specifying either undefined or specific prior distributions for any variable, and means for displaying or suppressing the display of the distribution of a variable when it is the same as the prior distribution.

13. The system of claim 1 further comprising an interface between an end user and the variables in the knowledge base, said interface comprising:

means for allowing the end user to override the computed value of a dependent variable; and

means for restoring the overridden variable to its unoverridden state.

14. The system of claim 13, said interface further comprising means for specifying a variable as nonoverrideable.

15. The system of claim 1 further comprising means for having more than one type of probability distribution for the values of a variable, with the ability for the distribution type to change dynamically at run time.

16. The system of claim 1 wherein said variables may be numeric or symbolic.

17. The system of claim 1 further comprising an interface between an end user and the variables in the knowledge base, said interface comprising:

means for specifying a set of display screens accessible to the end user as a function of the states of certain variables in the knowledge base.

18. The system of claim 17, said propagating means comprising means for minimizing the computation required during propagation by limiting the propagation to the subset of the dependent variables that are displayed to the end user.

19. The system of claim 18, said propagating means further comprising means for forcing propagation to extend beyond the displayed subset of dependent variables.

20. The system of claim 17 further comprising means for specifying special messages and the knowledge base conditions under which these special messages will be presented to an end user.

21. The system of claim 17 further comprising means responsive to an end-user request for clarifying the meaning of any displayed variable.

22. The system of claim 17, said interface further comprising means for specifying the layout, contents and appearance of each of the specified screens as a function of the states of specified variables in the knowledge base.

23. The system of claim 22, said interface further comprising means for specifying a set of variables whose values are displayed to the end user on said specified screens.

24. The system of claim 23, said interface further comprising means for specifying a set of variables whose values can be entered or changed by the end user.

25. The system of claim 24 said interface further comprising:

means for allowing the end user to override the computed value of a dependent variable; and

means for restoring the overridden variable to its unoverridden state.

26. The system of claim 25, said interface further comprising means for specifying a variable as nonoverridable.

2. The system of claim I further comprising means for creating and deleting instances of variables, either statically at compile time or dynamcially at run time.

27. An expert system shell comprising:

(a) a computer having a memory for storing a knowledge base, display means for displaying data, and entry means for entering data;

said knowledge base comprising;

(1) variables having a values represented by tables of probability distributions keyed by zero, one or more formal parameters;

(2) functions defining relationships between the values of each dependent variable and the values of its corresponding argument variables; and

(3) means for computing the probability distribution of the values of a dependent variable from the probability distribution of the values of the corresponding argument variables;

(b) means for propagating the consequences of a change in the value of any variable to maintain the defined functional relationships among a selected subset of the dependent variables; and

(c) an interface between the end user and the variables in the knowledge base, said interface comprising;

(1) means for allowing the end user to override the computed values of a dependent variable;

(2) means for restoring the overridden variable to its unoverridden state;

(3) means for specifying a set of screens accessible to the end user, and the layout, contents and appearance of the specified screens, as a function of the states of certain variables in the knowledge base;

(4) means for specifying a set of variables whose values are displayed to the end user on said specified screens; and

(5) means for specifying a set of variables whose values can be entered or changed by the end user.

28. The system of claim 27, said propagating means comprising means for limiting the propagation to the subset of the dependent variables that are displayed to the end user, thereby minimizing the computation required during propagation.

29. The system of claim 28, said propagating means further comprising means for forcing propagation to extend beyond the displayed subset of dependent variables.

30. The system of claim 27, said interface further comprising means for specifying a variable as nonoverridable.

31. The system of claim 27 whereas said variables may be numeric or symbolic.