Self-optimizing method and machine

• US 5,410,634 A
• Filed: 10/21/1991
• Issued: 04/25/1995
• Est. Priority Date: 09/19/1984
• Status: Expired due to Term

• Current Assignee: Chou H. Li
• Original Assignee: Chou H. Li
• Inventors: Li, Chou H.
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US07/767,154
• Time in Patent Office: 1,282 Days
• Field of Search: 395/13, 395/10
• US Class Current: 706/62
• CPC Class Codes: G05B 13/0235 using steepest descent or ascent method
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A method for automatically developing new and improved software according to a prespecified performance criterion comprising: automatically planning experiment on selected control variables related to the software; executing the planned experiment; analyzing the experimental results to compare the performance characteristics of the various combinations of the selected control variables in order to determine that combination which improves or optimizes the prespecified optimizing criterion; coding the thus-determined combination in computer-usable form; and saving the coded computer software onto a software storage device. A machine which automatically develops new, self-optimized software according to this method is also disclosed.

Claims

1. A method for automatically developing software according to a prespecified software performance in response to variations on a prescribed number m of selected control variables related to the software performance, m being an integer which is at least 7, comprising:

automatically planning a software experiment involving a set of n statistically designed software tests for testing in said n tests said m selected control variables, n being an integer greater than m, said n tests being centered around a specified experimental point in an m-dimensional experimental space;

performing said n designed software tests to collect experimental software performance results;

determining from said n experimental results the combination of said m variables which optimizes said prespecified software performance;

setting and coding in the developed software the conditions of said m variables at the thus-determined optimum variables combination; and

saving the thus-developed software in a software storage device.

2. A method as in claim 1 including the additional step of coupling the automatic planning, performing, determining, setting and coding, and saving steps for recycling said automatic software experiment to achieve self-optimized software.

3. A method as in claim 2 wherein except for the first software experiment cycle, said planning step comprises automatically replanning a new designed experiment centered around the optimizing variables combination in the immediately previous cycle.

4. A method as in claim 2 for operation without human performing at least one of the tasks selected from the group consisting of assigning a value to each test result, deciding selection criteria for choosing test samples or entities according to the assigned values, and choosing an operation to be performed during the software development.

5. A method as in claim 2 wherein said planning step comprises automatically planning said n designed tests in which each of said m selected control variables is systematically, not randomly, varied in all the n tests.

6. A method as in claim 2 wherein said automatic planning, performing, determining, setting and coding, and saving steps are performed with the only human knowledge base required being the number, location in the software, and permitted ranges of the m control variables.

7. A method as in claim 2 wherein said planning step comprises automatically planning the statistically designed, n-test software experiment in the form of a fractional factorial type.

8. A method as in claim 2 wherein said determining step comprises climbing along the steepest ascent path on a m-dimensional performance response surface in the search for the optimizing variables combination.

9. A method as in claim 2 wherein said determining step comprises systematically, fractionally but uniformly searching for said optimizing variables combination within the entire m-dimensional experimental space.

10. A method as in claim 2 wherein said automatic planning step comprises selecting a value of n which is no more than one-sixteenth of the required number to completely cover all the nodes in the m-dimensional experimental space.

11. A method as in claim 2 including the additional step of selecting a single computing system to perform said automatic planning, performing, determining, setting and coding, and saving steps.

12. A method as in claim 11 wherein said selecting step comprises selecting a personal computer as said single computing system.

13. A method as in claim 2 for developing the software to be used for or training educating a living object, wherein said selecting step comprises selecting a multimedia computer, and wherein said control variables are selected from the group consisting of type, intensity, and tempo of sound or noise, smell, lighting, temperature, humidity, pressure, visual display, and other factors that stimulate the object physically, mentally, psychologically, and physiologically, and wherein said performance relates to the physical, mental, psychological, or physiological condition of the object.

14. A method as in claim 2 for developing the software to be used in improving a living object, wherein said control variables are selected from the group consisting of type, intensity, and tempo of sound or noise, smell, lighting, temperature, humidity, pressure, visual display, and other factors that stimulate the object physically, mentally, psychologically, and physiologically, and wherein said performance criterion relates to the physical, mental, psychological, or physiological condition of the object.

15. A method as in claim 2 for developing the software to be used in cases where interactions exist among said selected control variables and wherein said determining step comprises automatically analyzing these interactions in the determination of the optimizing variables combination.

16. A method as in claim 2 for developing the software to be of a type selected from the group consisting of computer operating software and computer application software.

17. A method as in claim 2 wherein said prespecified software performance relates to at least one software feature selected from the group consisting of cost of development, productivity, smallness, accuracy, efficiency, utility, reliability, robustness, maintainability, flexibility, memory requirement, output quality, and computing or analyzing speed.

18. A method as in claim 2 including selecting said software control variables from the group consisting of number, location, initial point, ending point, and step size of at least one do loop in the software, number, type, and location of subroutines, and relational or logical operators.

19. A method as in claim 2 for developing the software to be used in improving or optimizing a specific growth performance of a growing object or creature and wherein said control variables are growth-controlling variables selected from the group consisting of equipment, materials, parts, procedures, and environment.

20. A machine for the automatic development of software according to a prespecified software performance in response to variations on a prescribed number m of selected control variables related to the software performance, m being an integer which is at least 7, said development being accomplished through computerized automatic software experimentation for systematically or non-randomly and uniformly searching for an optimizing variables combination in an m-dimensional experimental space, comprising:

means for automatically planning a software experiment involving a set of n statistically designed software tests for searching in said m-dimensional space, n being an integer greater than m, said n tests being centered around a specified experimental point in said m-dimensional space;

means for performing said n designed software tests to collect experimental software performance results for these tests;

means for determining from said n experimental results that combination of said m variables which optimizes said prespecified software performance;

means for setting and coding in the developed software the conditions of said m variables at the thus-determined optimum variables combination; and

means for saving the thus-developed software in a software storage device.

21. A machine as in claim 20 including means for coupling the planning, performing, determining, setting and coding, and saving means for recycling with feedback control to achieve the self-optimized software through the use of multiple, recycled optimizing designed experiments.

22. A machine as in claim 21 wherein except for the first optimizing software experiment said planning means comprises means for replanning a new designed experiment centered around the optimizing variables combination in the immediately previous cycle.

23. A machine as in claim 21 in which the only human knowledge base required for its operation is the number, location in the software, and permitted ranges of the m control variables.

24. A machine as in claim 21 wherein said planning means comprises means for automatically planning said n designed tests in which each of said m selected control variables is systematically, not randomly, varied in all the n tests.

25. A machine as in claim 21 wherein said planning means comprises means for automatically planning plans the statistically designed, n-test software experiment in the form of a fractional factorial type.

26. A machine as in claim 21 wherein said determining means comprises means for climbing the steepest ascent path on an m-dimensional performance response surface.

27. A machine as in claim 21 wherein n is no more than one-sixteenth of the required number to completely cover all the nodes in said m-dimensional space.

28. A machine as in claim 21 including a single computing system to provide said planning, performing, determining, setting and coding, and saving means.

29. A machine as in claim 28 wherein said single computing system is a personal computer.

30. A machine as in claim 21 for developing the software to be used for educating a living object, wherein said single computing system is a multimedia computer and wherein said selected control variables are selected from the group consisting of type, intensity, and tempo of sound or noise, smell, lighting, temperature, humidity, pressure, visual display, and other factors that stimulate the object physically, mentally, psychologically, and physiologically, and wherein said performance relates to the physical, mental, psychological, or physiological condition of the object.

31. A machine as in claim 21 for uses in cases where interactions exist among said selected control variables and wherein said determining means comprises means for automatically analyzing these interactions in the determination of the optimizing variables combination.

32. A machine as in claim 21 wherein said software to be developed is of a type selected from the group consisting of computer operating software and computer application software.

33. A machine as in claim 21 wherein said prespecified software performance relates to at least one software feature selected from the group consisting of cost of development, productivity, smallness, accuracy, efficiency, utility, reliability, robustness, maintainability, flexibility, memory requirement, output quality, and computing or analyzing speed.

34. A machine as in claim 21 wherein said software control variables are selected from the group consisting of number, location, initial point, ending point, and step size of at least one do loops in the software, number, type, and location of subroutines and relational or logical operators.

35. A method for automatically developing, in real time, software relative to a specific performance of the software to be improved or optimized comprising:

performing at least one automatic, computerized software experiment involving a plurality of selected control variables which affect the specific performance;

computer-analyzing, in real time, the experimental results to determine the best combination of the selected control variables for achieving computerized optimization or the most known improvement at the time of said computerized experiment;

setting and coding in the developed software the conditions of said selected control variables at the thus-determined optimum or most known improved variables combination; and

saving the thus-generated software in a software storage device.

36. A method as in claim 35 including the additional steps of:

instantly feeding back information on the status of improvement or optimization of said specific performance;

reperforming at least one more similar automatic computerized software experiment; and

computer-reanalyzing, in real time, the new experimental results to acquire newer and better improvement or optimization;

resetting and recoding in the developed software the new conditions of said variables at the thus-determined, new optimum or most known improved variables combination; and

resaving the thus-generated improved software in the software storage device.

37. A method as in claim 36 wherein said reperforming step after the first computerized experiment comprises performing said subsequent at least one computerized experiment centered around the best variables combination determined in the immediately previous computerized experiment.

38. A method as in claim 37 wherein the number of said selected control variables is at least 7.

39. A method as in claim 37 wherein each of said automatic, computerized software experiments comprising said performing or reperforming step, computer-analyzing or computer-reanalyzing step, setting and coding or resetting and recoding step, and saving or resaving step is finished within a minute.

40. A method as in claim 37 wherein said computer-analyzing or reanalyzing step comprises computer-analyzing or reanalyzing the experimental results within a time selected from the group consisting of 1.79 milliseconds and 1.79 microseconds.

41. A method as in claim 37 wherein said computer-analyzing and reanalyzing steps comprise computer-determining the effect, at different times of the successive automatic computerized experiments, of at least one of said selected control variables to achieve a serial time effect of the at least one selected control variable.

42. A method as in claim 37 wherein all the automatic computerized software experiments are performed without any human involvement in a form selected from the group consisting of human control, human supervision, human guidance, and supply of human knowledge bases other than providing the number and type of said selected control variables, and their allowable ranges.

43. A method as in claim 37 wherein said software is a multi-media software.

44. A method as in claim 37 for developing software used to improve the physical or mental condition of a living object;

wherein said control variables are selected from the group consisting of the sex, age, and background of the object;

time of the day or week;

video and audio signals including those from recorded tapes;

environmental factors including lighting, temperature, humidity, smell, noise, and pressure testing material or equipment; and

encouragement rewards; and

wherein said specific performance is selected from the group consisting of the speed or cost of improvement and maximum improvement in the physical or mental condition of the object.