Self-optimization with interactions

• US 6,513,024 B1
• Filed: 03/16/1999
• Issued: 01/28/2003
• Est. Priority Date: 03/16/1999
• Status: Expired due to Fees

• Current Assignee: LI Family Holding Limited
• Original Assignee: LI Family Holding Limited
• Inventors: Li, Chou H.
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US09/268,297
• Time in Patent Office: 1,414 Days
• Field of Search: 706/45, 706/13, 706/25
• US Class Current: 706/45
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A method for computer-generating interaction-specific knowledge base for rapidly improving or optimizing a performance of an object comprises performing, according to computer-designed test matrices, at least several automatic experimental cycles on selected control variables. In at least one of the automatic experimental cycles after the first the computer plans a new test matrix designed to minimize or remove at least one expected two-variable interaction from a main effect of a designated control variable. A machine operating according to the method is also available.

Claims

1. An article of manufacture comprising:

a computer usable medium having computer readable program code means embodied therein for self-optimizing or self-improving a performance of a computer optimizable object having a plurality of selected control variables, the computer readable program code means in said article of manufacture comprising;

computer readable program code means for performing a plurality of computerized automatic experimental cycles on said computer optimizable object relative to said plurality of selected control variables, said computer readable program code means performing the steps of;

computer-planning a designed experiment for each of the plurality of the computerized automatic experimental cycles;

said computer-planning step comprising computer-planning at least one new computer designed experiment in a new experimental region not centered around a point representing an optimum variable combination in an immediately previous experimental cycle;

computer-executing each experimental cycle according to the respective designed experiment to obtain test results;

computer-analyzing, in real time, the test results to determine a way to combine optimally the selected control variables for optimizing or improving the performance of the computer optimizable object; and

computer-coding the computer usable data structure for storage in a selected computer readable data storage means; and

computer-storing the computer coded and computer analyzed data structure in the selected computer readable data storage means.

2. An article of manufacture according to claim 1 wherein prior to said computerized automatic experimental cycles, said performance is selected by a human or machine automation manager;

and

after at least one of said computerized automatic experimental cycles said human or machine automation manager decides on whether or which two-variable interaction is to be further analyzed.

3. An article of manufacture according to claim 2 wherein said computer-planning step further comprises computer-planning said computer-designed experiment to treat a minimum number of control variables selected from a group consisting of 15, 31, 63, 127, 255, 511, and 1023.

4. An article of manufacture according to claim 2 wherein said computer-planning step further comprises computer-planning the designed experiment for separating the main effect of the control variables from selected combined interaction effects of multiple variables.

5. An article of manufacture according to claim 2 wherein said performance relates to a computer self-optimizable physical or mental growth activity of a living object selected from the group consisting of human, animal, bacteria, virus, germ, and plant.

6. An article of manufacture as in claim 1 wherein said computer-planning step comprises computer-planning for the at least one experimental cycle after the first cycle, a new computer-designed experiment to estimate at least one expected two-variable interaction effect in addition to main effects of at least two designated control variables;

a main effect being a computer-analyzed individual effect of a single control variable on the performance of the object, while a two-variable interaction effect being a computer-analyzed combined effect of two selected control variables on the performance.

7. A computer program product for use in self-optimizing or self-improving a performance of a computer optimizable object having a plurality of selected control variables, said computer program product comprising:

a computer usable medium having computer readable program code means embodied in said medium for performing a plurality of computerized automatic experimental cycles on said plurality of selected control variables, said computer program product having;

computer readable program code means for causing a computer to self-plan a designed experiment for each of the plurality of the computerized automatic experimental cycles;

said computer self-planning step comprising;

computer-planning, after the first computerized automatic experimental cycle, at least one new computer-designed experiment for an immediately following experimental cycle;

computer readable program code means for causing a computer to execute the experimental cycles according to the respective computer-planned designed experiments to obtain test results;

computer readable program code means for causing a computer to computer-analyze, in real time, the test results, to find a way to combine optimally the selected control variables for optimizing or improving the performance of the object;

computer readable program code means for causing the computer to computer-transform the computer-analyzed test results into a computer-usable data structure;

computer-readable program code means for causing the computer to computer-code the computer-usable data structure for storage; and

computer readable program code means for causing the computer to computer-store the computer-analyzed, computer-transformed, and computer-coded data structure in a selected computer readable data storage device.

8. A computer program product according to claim 7 wherein said object is a computer self-optimizable or self-improvable object;

a performance of said object being selected by a human or machine automation manager for computer self-optimization or self-improvement to accomplish a selected task in practical optimization technology for a specific industry.

9. A computer program product according to claim 7 wherein said computer program product further comprises computer readable program code means embodied in said medium for computer-transforming said computer-analyzed test results into a computer-usable data structure.

10. A computer program product as in claim 7 wherein said self-planning computer readable program code means comprises means for computer-planning, for at least one of the experimental cycles after the first cycle, a new computer-designed experiment in a new experimental region having a center different from that in a previous experimental region, to estimate at least one, likely important two-variable interaction effect in addition to main effects of these two selected control variables to thereby generate an interaction-specific knowledge base in a specific domain of technology;

a main effect being a computer-analyzed individual effect of a single control variable on the performance of the object, while a two-variable interaction effect being a computer-analyzed combined effect of two selected control variables on the performance.

11. A program storage device readable by a machine, tangibly embodying a program of instructions executable by the machine to perform method steps for a plurality of automatic computerized experimental cycles according to a test matrix each involving a plurality of selected control variables capable of use thereby self-optimizing or self-improving a performance of a computer optimizable object having a plurality of selected control variables, said method steps comprising:

computer-planning a designed experiment for each one of the plurality of automatic computerized experimental cycles;

the computer designed test matrix in one experimental cycle being different from that of at least another experimental cycle,

computer-executing each experimental cycle according to the respective computer-planned designed experiment to obtain test results;

computer-analyzing, in real time, the test results for the experimental cycle to determine a way to combine optimally said plurality of selected control variables for optimizing or improving the performance of the object;

computer-transforming the computer-analyzed test results into a specific computer-transformed data structure; and

computer-coding the specific computer-transformed data structure for storage in a selected computer readable data storage means.

12. A program storage device according to claim 11 wherein said method steps further comprise separating or purifying at least one main effect of said plurality of selected control variables from a selected combined interaction effect of multiple variables.

13. A program storage device as in claim 11 wherein said method steps comprise:

computer planning, for at least one of said experimental cycles after the first cycle, a new computer-designed experiment to estimate at least one expected two-variable interaction effect in addition to main effects of at least two designated control variables;

a main effect being a computer-analyzed individual effect of a single control variable on the performance of the object, while a two-variable interaction effect being a computer-analyzed combined effect of two selected control variables on the performance.

14. A memory for storing data for access by an applied optimization program used in a specific industry and being executed on a data processing system for optimizing a computer optimizable object, comprising:

a data structure stored in said memory, said data structure including information on a plurality of selected control variables resident in a knowledge base used by said application program;

said knowledge base being real-time and computer-generated, and containing at least an estimate of an expected two-variable interaction effect in addition to main effects of at least two designated control variables;

a main effect being a computer-analyzed individual effect of a single control variable on the performance of the object, while a two-variable interaction effect being a computer-analyzed combined effect of two selected control variables on the performance.

15. A memory for storing data for access by an applied optimization program used in a specific industry and being executed on a data processing system for optimizing a computer optimizable object, comprising:

a data structure stored in said memory, said data structure including information on a plurality of selected control variables resident in a knowledge base used by said application program;

said knowledge base comprising a real-time and computer-generated, interaction-variable-specific knowledge base.

16. A machine having a memory device which contains the knowledge base computer-generated by the method of claim 15, 17, 19, 23, 24, 28, or 30.

17. A data processing system executing an optimization application program and containing a computer-generated knowledge base used by said application program to perform a practical optimization task in a specific industry;

said data processing system comprising;

a computer device for processing said application program; and

a memory component for holding said computer-generated knowledge base for access by said application program, said knowledge base being composed of information resident in said knowledge base used by said application program and including;

determining a presence or absence of a suspected two-variable interaction effect in addition to main effects of at least two designated control variables;

a main effect being a computer-analyzed individual effect of a single control variable on the performance of a computer optimizable object, while a two-variable interaction effect being a computer-analyzed combined effect of two selected control variables on the performance.

18. A computer-readable medium containing instructions for self-optimizing or self-improving a performance of a computer self-improvable object relative to a plurality of control variables by:

computer-planning a plurality of computerized automatic experimental cycles each containing a computer-designed test matrix;

computer-executing the experimental cycles according to the respective computer designed test matrices to collect test results;

computer-analyzing, in real time, the test results to determine a way to combine optimally the selected control variables for computer optimizing or improving the performance of the computer self-improvable object;

computer-coding the computer-analyzed test results for storing a selected computer readable program code means; and

when it is determined that the performance of said computer self-improvable object is substantially optimized or improved, recording the optimal combination of the control variables and keeping said computer self-improvable object in the thus-optimized or improved condition.

19. A computer-readable data transmission medium containing a real-time computer-generated knowledge base for optimizing a computer optimizable object and generated by automatic computerized experimentation and optimization in at least two different experimental regions;

said real-time computer-generated knowledge base including an optimal combination of selected control variables in a specific field of technology.

20. A computer-readable data transmission medium according to claim 19 wherein said real-time computer-generated knowledge base is an interaction-specific knowledge base giving an optimized or improved effect on a physical, mental, medical, psychological, or physiological condition or activity of a living computer optimizable object.

21. A computer-readable data transmission medium according to claim 19 wherein said real-time computer-generated knowledge base comprises an instantly machine-generated, interaction-variable identification knowledge base.

22. A computer-readable data transmission medium according to claim 19 wherein said real-time computer-generated knowledge base comprises an interaction-variable identification knowledge base to keep all the main effects and at least one selected two-variable interaction substantially pure.

23. A method in a computer system for communicating with an automation system, the method comprising:

real-time automatic computerized experimenting in at least two different experimental regions on a plurality of selected control variables in at least one computer-designed experiment;

collecting the test results;

computer-analyzing the test results to generate a new knowledge base including at least an optimal combination of the control variables;

computer-coding in real time the new knowledge base into a machine-usable form; and

in response to the automation system, telecommunicating, in real time, the computer-coded, new knowledge base to the automation system for its immediate implementation to achieve, instantly and without additional automatic computerized experimentation, a substantially optimal condition thereon.

24. A computer-readable data transmission medium containing a knowledge base, comprising:

real-time computer-generated knowledge base including an optimal combination of selected control variables in a specific field of technology for optimizing a computer optimizable object;

said knowledge base having at least one estimate of an expected two-variable interaction effect in addition to main effects of two designated control variables;

a main effect being a computer-analyzed individual effect of a single control variable on the performance of the computer optimizable object, while a two-variable interaction effect being a computer-analyzed combined effect of two selected control variables on the performance.

25. A computer-readable data transmission medium according to claim 24 wherein said knowledge base is computer-generated using an interaction-variable identification knowledge base for finding at least two of the following:

a) the most important variables;

b) their main effects;

c) if interaction effects exist, at least one important two-variable interaction effect;

d) the best way to isolate the control variables from the interaction effects between variables to insure minimum cross-contamination of the main effects and interaction effect.

26. A computer-readable data transmission medium according to claim 24 wherein said knowledge base is computer-generated by running a computer-designed experiment not around a previous optimum variable combination, but at a tolerable distance away from this optimum variable combination, to search for a better global optimum rather than a local optimum,

the tolerable distance being such that it is still within the production specification.

27. A computer-readable data transmission medium according to claim 24 wherein said computer-generated knowledge base has at least two of the following features:

inexpensive, relevant, timely, unbiased, and reliable.

28. A memory for storing a knowledge base for access by an application program being executed on a data processing system for optimizing a computer optimizable object, comprising:

a knowledge base real-time computer-generated based on automatic computerized testing in at least two different experimental regions and stored in said memory, and including a computer-coded, new database for direct and immediate use by said application program.

29. A method for generating a knowledge base for access by an application program executable by a data processor, said method comprising the method steps, executed by said data processor, of:

a) computer-planning a designed experiment for each of a plurality of computerized automatic experimental cycles on a plurality of selected control variables;

said designed experiment being computer-planned to include at least 7 or 15 control variables with no more than one-sixteenth of a number of experimental tests normally required for a complete factorial experimentation, for determining presence or absence of at least one interaction effect of two selected control variables;

said computer-planning step comprising using an interaction-variable identification knowledge base to cause the computer to automatically design the new design matrices specifically to isolate a most important main effect from at least one two-variable interaction effect;

b) computer-executing the experimental cycles according to the respective computer-planned designed experiments to obtain test results;

c) computer-analyzing, in real time, the test results to determine at least a way to combine the selected control variables for optimizing a performance of a computer optimizable object;

said computer-analyzing step additionally determining presence or absence of at least one suspected two-variable interaction effect in addition to substantially pure main effects of at least two designated control variables, a main effect being a computer-analyzed individual effect of a single control variable on the performance of the computer optimizable object, and a two-variable interaction effect being a computer-analyzed combined effects of two selected control variables on the performance of the object; and

d) computer-coding an interaction-specific knowledge base generated by said computer-analyzing step for storage in a selected computer readable program memory means.

30. A method for self-optimizing or self-improving a performance of a computer optimizable object for use by an application program being executable by a data processor, said method comprising the method steps, executed by said data processor, of:

a) planning with a computer a designed experiment for each of a plurality of computerized automatic experimental cycles on a plurality of selected control variables;

b) sequentially computer-executing the experimental cycles according to the respective computer-planned designed experiments to obtain test results;

c) computer-analyzing, in real time, the test results to computer-generate an interaction-specific knowledge base relating to presence or absence of at least one combined effect of at least two selected control variables and also to a way for optimizing or improving the performance of said object;

d) computer-coding the computer-generated interaction specific knowledge for storage in a selected computer memory device; and

e) computer-telecommunicating, in real time, said computer-generated interaction specific knowledge base to a computing means for use not in starting its own new computerized automation experimentation, but as an embedded expert system that performs at least a plurality of the following;

aa) understands the computer-generated and computer-coded knowledge base;

bb) logically compares different parts of the knowledge base;

cc) selects relevant compared parts of the knowledge base;

dd) synthesizes new method steps, based on logical decisions, for further improving the performance of the computer optimizable object; and

ee) provides or prescribes a new synthesized recipe for achieving an improved performance of the computer optimizable object with the use of said computing means.

31. A method according to claim 30 wherein said computer-generated interaction specific knowledge base is substantially instantly machine generated.

32. In a data processing system executing a self-optimization application program and containing a computer generated knowledge base used by said application program to perform, in a plurality of computerized automatic experimental cycles on a computer self-optimizable object, a practical self-optimization task in a specific industry;

said data processing comprising;

a computer device for processing said application program; and

a memory component for holding said computer generated knowledge base for access by said application program, said knowledge base being comprised of information resident in said knowledge base used by said application program, a method of computer self-optimizing said object comprising the steps of;

computer-planning a designed experiment for each of the plurality of the computerized automatic experimental cycles;

at least one of the new designed experiments in one computerized automatic experimental cycle differing from another designed in another computerized automatic experimental cycle in a feature selected from the group consisting of;

a) a center of the experimental region;

b) a sequence of the selected controlled variables in a design test matrices;

computer-executing each experimental cycle according to the respective designed experiment to obtain test results;

computer-analyzing, in real time, the test results to determine a way to combine optimally the selected control variables for optimizing the performance of the computer optimizable object; and

computer-coding the computer usable data structure for storage in a selected computer readable data storage means.

33. A machine for self-optimizing or self-improving a performance of a computer optimizable object having a plurality of selected control variables comprising:

computer means for performing at least one computerized automatic experiment on said plurality of selected control variables to computer-generate a knowledge base associated with the performance; and

means for performing at least a plurality of the following tasks;

a) self-learning, self-improving, and self-optimizing;

b) understanding, by a computer, the computer-generated knowledge base associated with the performance of the object;

c) logically and by a computer, comparing different parts of the knowledge base;

d) computer-selecting relevant different parts of the knowledge base;

e) computer-synthesizing at least one new method step based on a logical decision associated with at least a part of the knowledge base;

f) providing or prescribing a new computer-synthesized recipe for achieving an improved performance on the computer optimizable object with the use of a processor; and

g) telecommunicating to a computer device a computer synthesized recipe or computer-generated knowledge base to achieve instantly an improved performance without additional computerized automatic experimentation on the object.

34. The machine of claim 33 in which the computer-generated knowledge base has at least a plurality of the following features:

a) generated by the self-optimizing or self-improving machine experimenting in real-time on the very computer optimizable object whose performance is to be improved;

b) either interaction specific or interaction non-specific; and

c) immediately implementable on a computer device to achieve instant improved performance on the object.