Method, system, and program for converting code to executable code using neural networks implemented in a software program

• US 6,832,214 B1
• Filed: 12/07/1999
• Issued: 12/14/2004
• Est. Priority Date: 12/07/1999
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Nguyen, Chung T.
• Primary Examiner(s): Knight, Anthony
• Assistant Examiner(s): Booker, Kelvin
• Application Number:
  US09/455,704
• Time in Patent Office: 1,834 Days
• Field of Search: 706/15, 706/27, 706/26, 706/20, 706/30, 717/139, 717/148, 717/138, 717/106, 717/108, 717/114, 717/116, 717/118, 717/136, 717/137, 717/140, 717/146, 717/147
• US Class Current: 706/15
• CPC Class Codes: G06F 8/41 Compilation

Abstract

Disclosed is a system, method, and program for generating a compiler to map a code set to object code capable of being executed on an operating system platform. At least one neural network is trained to convert the code set to object code. The at least one trained neural network can then be used to convert the code set to the object code.

Claims

1. A method for generating a compiler to map a code set to object code capable of being executed on an operating system platform, comprising:

training at least one neural network to convert the code set to object code, wherein the code set comprises an intermediary platform independent code generated by processing source code; and

using the at least one trained neural network to convert the code set to object code.

2. The method of claim 1, wherein the code set comprises Java bytecode implementing a Java class.

3. The method of claim 1, wherein the code set represents an implementation of an object oriented class.

4. The method of claim 1, wherein training the at least one neural network comprises:

training at least one first level neural network to convert the code set to a unique representation of the code set; and

training a second level neural network to convert the unique representation of the code set to the object code.

5. The method of claim 4, wherein training the at least one first level neural network comprises:

parsing the code set into a plurality of input vectors; and

assigning each input vector to one of a plurality of first level neural networks, wherein each of the first level neural networks are trained to convert the assigned input vector to an output vector, wherein the combination of the output vectors from each first level neural network forms a unique representation of the code set.

6. The method of claim 5, wherein the code set comprises a Java bytecode implementing a Java class, and wherein parsing the code set into a plurality of input vectors, comprises parsing the code set into categories based on the bytecode class file format, wherein each input vector includes bits from the bytecode that comprise one of the categories.

7. The method of claim 5, wherein the second level neural network comprises a recurrent neural network, and wherein training the second level neural network comprises performing iterations inputting the unique representation of the code set to the second level neural network to generate the object code, wherein the generated object code is fed back into the second level neural network until the second level neural network is capable of mapping the unique representation of the code set to the object code with a minimal amount of error in the mapping.

8. The method of claim 5, wherein the first level neural networks comprise a self-organizing map neural network.

9. The method of claim 1, further comprising implementing the at least one neural network in a software program that is capable of mapping the input code set to the object code.

10. The method of claim 1, wherein the software program is implemented in a Java Just-In-Time Compiler component of a Java Virtual Machine.

11. The method of claim 1, wherein converting with the at least one neural network the received code set to the object code further comprises:

mapping, with each of a plurality of neural networks, a predefined section of the code set to an output vector;

combining the output vectors from the neural networks to form a combined output vector; and

processing the combined output vector to produce the object code.

12. A method for mapping a code set to object code capable of being executed on an operating system platform, comprising:

receiving the code set with at least one neural network, wherein the code set comprises an intermediary platform independent code generated by processing source code; and

converting with the at least one neural network the received code set to the object code.

13. The method of claim 12, wherein converting, with the at least one neural network, the code set to the object code comprises:

converting, with at least one first level neural network, the code set to a unique representation of the code set; and

converting, with a second level neural network, the unique representation of the code set to the object code.

14. The method of claim 13, wherein converting the code set to the unique representation with the at least one first level neural network comprises:

parsing the code set into a plurality of input vectors;

assigning each input vector to one of a plurality of first level neural networks;

converting, with each of the first level neural networks, the assigned input vector to an output vector; and

combining the output vectors from each first level neural network to form a unique representation of the code set.

15. The method of claim 12, wherein the code set comprises Java bytecode implementing a Java class and the at least one neural network is implemented in a Just-In-Time Compiler portion of a Java Virtual Machine.

16. A system for generating a compiler to map a code set to object code capable of being executed on an operating system platform, comprising:

means for training at least one neural network to convert the code set to object code, wherein the code set comprises an intermediary platform independent code generated by processing source code; and

means for using the at least one trained neural network to convert the code set to object code.

17. The system of claim 16, wherein the code set comprises Java bytecode implementing a Java class.

18. The system of claim 16, wherein the code set represents an implementation of an object oriented class.

19. The system of claim 16, wherein the Means for training the at least one neural network comprises:

means for training at least one first level neural network to convert the code set to a unique representation of the code set; and

means for training a second level neural network to convert the unique representation of the code set to the object code.

20. The system of claim 19, wherein the means for training the at least one first level neural network comprises:

means for parsing the code set into a plurality of input vectors; and

means for assigning each input vector to one of a plurality of first level neural networks, wherein each of the first level neural networks are trained to convert the assigned input vector to an output vector, wherein the combination of the output vectors from each first level neural network forms a unique representation of the code set.

21. The system of claim 20, wherein the code set comprises a Java bytecode implementing a Java class, and wherein the means for parsing the code set into a plurality of input vectors, comprises parsing the code set into categories based on the bytecode class file format, wherein each input vector includes bits from the bytecode that comprise one of the categories.

22. The system of claim 21, wherein the second level neural network comprises a recurrent neural network, and wherein the means for training the second level neural network comprises means for performing iterations inputting the unique representation of tic code set to the second level neural network to generate the object code, wherein the generated object code is fed back into the second level neural network until the second level neural network is capable of mapping the unique representation of the code set to the object code with a minimal amount of error in the mapping.

23. The system of claim 19, wherein the first level neural networks comprise a self-organizing map neural network.

24. The system of claim 16, further comprising means for implementing the at least one neural network in a software program that is capable of mapping the input code set to the object code.

25. The system of claim 24, wherein the software program is implemented in a Java Just-In-Time Compiler component of a Java Virtual Machine.

26. A system for mapping a code set to object code capable of being executed on an operating system platform, comprising:

means for receiving the code set, wherein the code set comprises an intermediary platform independent code generated by processing source code; and

neural network means for converting the code set to the object code.

27. The system of claim 26, wherein the neural network means further comprises:

first level neural network means for converting the code set to a unique representation of the code set; and

second level neural network means for converting the unique representation of the code set to the object code.

28. The system of claim 27, wherein the first level neural network means further comprising:

means for parsing the code set into a plurality of input vectors; and

means for assigning each input vector to one of a plurality of first level neural networks;

means for converting the assigned input vector to an output vector; and

means for combining the output vectors from each first level neural network to form a unique representation of the code set.

29. The system of claim 26, wherein the code set comprises Java bytecode implementing a Java class and the at least one neural network is implemented in a Just-In-Time Compiler portion of a Java Virtual Machine.

30. The system of claim 26, wherein the neural network means for converting the received code set to the object code further performs:

mapping, with each of a plurality of neural networks, a predefined section of the code set to an output vector;

combining the output vectors from the neural networks to form a combined output vector; and

processing the combined output vector to produce the object code.

31. An article of manufacture for use in generating a compiler to map a code set to object code capable of being executed on an operating system platform, the article of manufacture comprising computer useable media accessible to a computer, wherein the computer usable media includes at least one computer program that is capable of causing the computer to perform:

training at least one neural network to convert the code set to object code, wherein the code set comprises an intermediary platform independent code generated by processing source code; and

using the at least one trained neural network to convert the code set to object code.

32. The article of manufacture of claim 31, wherein the code set comprises Java bytecode implementing a Java class.

33. The article of manufacture of claim 31, wherein the code set represents an implementation of an object oriented class.

34. The article of manufacture of claim 31, wherein training the at least one neural network comprises:

training at least one first level neural network to convert the code set to a unique representation of the code set; and

training a second level neural network to convert the unique representation of the code set to the object code.

35. The article of manufacture of claim 34, wherein training the at least one first level neural network comprises:

parsing the code set into a plurality of input vectors; and

assigning each input vector to one of a plurality of first level neural networks, wherein each of the first level neural networks are trained to convert the assigned input vector to an output vector, wherein the combination of the output vectors from each first level neural network forms a unique representation of the code set.

36. The article of manufacture of claim 35, wherein the code set comprises a Java bytecode implementing a Java class, and wherein parsing the code set into a plurality of input vectors, comprises parsing the code set into categories based on the bytecode class file format, wherein each input vector includes bits from the bytecode that comprise one of the categories.

37. The article of manufacture of claim 35, wherein the second level neural network comprises a recurrent neural network, and wherein mining the second level neural network comprises performing iterations inputting the unique representation of the code set to the second level neural network to generate the object code, wherein the generated object code is fed back into the second level neural network until the second level neural network is capable of mapping the unique representation of the code set to the object code with a minimal amount of error in the mapping.

38. The article of manufacture of claim 35, wherein the first level neural networks comprise a self-organizing map neural network.

39. The article of manufacture of claim 31, further comprising implementing the at least one neural network in a software program that is capable of mapping the input code set to the object code.

40. The article of manufacture of claim 31, wherein the software program is implemented in a Java Just-In-Time Compiler component of a Java Virtual Machine.

41. An article of manufacture for use in mapping a code set to object code capable of being executed on an operating system platform, the article of manufacture comprising computer useable media accessible to a computer, wherein the computer usable media includes at least one computer program that is capable of causing the computer to perform:

using at least one neural network to convert the code set to the object code, wherein the code set comprises a intermediary platform independent code generated by processing source code; and

converting with the at least one neural network the received code set to the object code.

42. The article of manufacture of claim 41, wherein converting, with the at least one neural network, the code set to the object code comprises:

converting, with at least one first level neural network, the code set to a unique representation of the code set; and

converting, with a second level neural network, the unique representation of the code set to the object code.

43. The article of manufacture of claim 42, wherein converting the code set to the unique representation with the at least one first level neural network comprises:

parsing the code set into a plurality of input vectors;

assigning each input vector to one of a plurality of first level neural networks;

converting, with each of the first level neural networks, the assigned input vector to an output vector; and

combining the output vectors from each first level neural network to form a unique representation of the code set.

44. The article of manufacture of claim 41, wherein the code set comprises Java bytecode implementing a Java class and the at least one neural network is implemented in a Just-In-Time Compiler portion of a Java Virtual Machine.

45. The article of manufacture of claim 41, wherein converting with the at least one neural network the received code set to the object code further comprises:

mapping, with each of a plurality of neural networks, a predefined section of the code set to an output vector;

combining the output vectors from the neural networks to form a combined output vector; and

processing the combined output vector to produce the object code.

46. A method for generating a compiler to map a code set to object code capable of being executed on an operating system platform, comprising:

training at least one neural network to convert the code set to object code by training each of a plurality of neural networks to map a predefined section of the code set to an output vector, wherein the output vectors from the neural networks are combined to form a combined output vector that is further processed to produce the object code; and

using the at least one trained neural network to convert the code set to object code.

47. A system for generating a compiler to map a code set to object code capable of being executed on an operating system platform, comprising:

means for training at least one neural network to convert the code set to object code by training each of a plurality of neural networks to map a predefined section of the code set to an output vector, wherein the output vectors from the neural networks are combined to form a combined output vector that is further processed to produce the object code; and

means for using the at least one trained neural network to convert the code set to object code.

48. An article of manufacture for use in generating a compiler to map a a code set to object code capable of being executed on an operating system platform, the article of manufacture comprising computer useable media accessible to a computer, wherein the computer usable media includes at least one computer program that is capable of causing the computer to perform:

training at least one neutral network to convert the code set to object code by training each of a plurality of neural networks to map a predefined section of the code set to an output vector, wherein the output vectors from the neural networks are combined to form a combined output vector that is further processed to produce the object code; and

using the at least one trained neural network to convert the code set to object code.