Self-programming artificial intelligence for generating machine code based on machine learning and statistical algorithm

Abstract

An artificial intelligence technology is widely applied to the fields of image recognition, voice recognition, text recognition and the like, but the artificial intelligence technology excessively depends on a neural network model and computing power resources, so that the maintenance difficulty is large, the operation cost is high, and an algorithm structure is difficult to flexibly design according to actual requirements in the domestic use process. The invention constructs a self-programming artificial intelligence system for generating a machine code based on machine learning and a statistical algorithm, the system collects a program source code and a corresponding assembly instruction, a machine code and a natural language based on a big data technology, and a mapping system among the natural language, a programming language, the assembly instruction and the machine code is constructed. And finally, an artificial intelligence system capable of self-compiling an executable program is established.

Description

Technical Field

[0001] The present invention belongs to the fields of artificial intelligence and computer, and particularly relates to a self-programming artificial intelligence that generates machine code based on machine learning and statistical algorithms. Technical Background

[0002] Recently, artificial intelligence and big data technologies have been booming in China. Among them, artificial intelligence technology has been widely used in fields such as image recognition, speech recognition, and text recognition, greatly improving the efficiency of practical production work. However, due to its excessive dependence on the large deep learning neural network model (this model is mainly constructed by the foreign Google team), there are difficulties in maintaining the underlying code during domestic use, and it is difficult to flexibly design and adjust the algorithm structure according to actual work requirements. Therefore, its application scope is still limited to the processing of large-scale image, speech, and text data.

[0003] The present invention constructs a self-programming artificial intelligence system that generates machine code based on machine learning and statistical algorithms. This system collects program source codes written in programming languages such as C / C++, C#, and Python, as well as their corresponding assembly instructions, machine codes, and natural language data based on big data technology. Then, through statistical models such as natural language processing models, machine learning models, generalized linear models, and simple linear models, the construction of mapping functions between "natural language - programming language - assembly instruction - machine code" is completed. Then, through loop iteration, the mapping system between natural language and machine code is backtested to screen and generate a mapping scheme for correct and efficient machine code. Finally, an artificial intelligence system that can self-write executable programs is established based on the optimal mapping set. Summary of the Invention

[0004] The present invention provides a self-programming artificial intelligence system that generates machine code based on machine learning and statistical algorithms, in order to reduce the code maintenance cost of the large deep learning neural network model, expand the application scope of artificial intelligence technology, and on this basis, further improve the program writing efficiency through the self-programming artificial intelligence system, and promote the development of domestic artificial intelligence and computer technologies.

[0005] The self-programming artificial intelligence system of the present invention that generates machine code based on machine learning and statistical algorithms includes the following steps:

[0006] 1. Natural language set: Based on big data technology, collect the English names and definitions of function bodies involved in programming languages such as C / C++, C#, and Python.

[0007] 2. Programming language set: Based on big data technology, collect the source codes of executable programs written in programming languages such as C / C++, C#, and Python.

[0008] 3. Assembly Instructions and Machine Code Set: Disassemble the executable program to collect the assembly instruction set and machine code set inside the program.

[0009] 4. Establish a Hierarchical Multidimensional Dataset: Organize each dataset into a hierarchical multidimensional dataset of "Natural Language - Programming Language - Assembly Instructions - Machine Code". Among them, the languages between different levels present a certain topological structure. For example, the same natural language can correspond to different programming languages and combinations of assembly instructions, and the same programming language can also correspond to different natural languages and combinations of assembly instructions. However, there is a one-to-one correspondence between assembly instructions and machine code, while there can be a multiple correspondence between combinations of assembly instructions and combinations of machine code.

[0010] 4. Construct Mapping Functions: Based on statistical models such as natural language processing models, machine learning models, generalized linear models, and simple linear models, construct mapping functions between languages layer by layer until the construction of the overall mapping function system of "Natural Language - Programming Language - Assembly Instructions - Machine Code" is completed. Among them, a natural language processing model is used to construct the mapping function between natural language and programming language, a generalized linear model is used to construct the mapping function between programming language and assembly instructions, and machine learning models and simple linear models are used to construct the mapping function between assembly instructions and machine code.

[0011] 5. Generate Machine Code from Natural Language: Send natural language instructions to the established mapping system to generate the corresponding machine code.

[0012] 6. Package Machine Code into an Executable Program: Write an automated program to implement the batch packaging of machine code into executable files.

[0013] 7. Test the Executable Program and Score the Intelligent System: Conduct a running test on the generated executable file, screen the executable programs that run normally and occupy less memory, and record this machine code generation scheme as a high-score scheme in the intelligent system.

[0014] 8. Iterate through the loop in steps 5 - 7 to generate an intelligent self-programming system with stable performance: Iteratively loop through the programming process of the intelligent system until it stably generates the optimal program. Description of the Drawings

[0015] Figure 1 Shows the partial correspondence between machine code and assembly instructions;

[0016] Figure 2 Is a schematic diagram of the mapping topological structure between languages at different levels;

[0017] Figure 3 Is a schematic diagram of the mapping path and main models from natural language to machine code;

[0018] Figure 4 Shows the partial machine code generated after inputting natural language instructions. Specific implementation mode

[0019] The following examples further illustrate the technical solutions of the present invention, but do not limit the content of the invention:

[0020] 1. Natural language set: Based on big data technology, collect the English names and definitions of function bodies involved in programming languages such as C / C++ language, C#, Python, etc.

[0021] 2. Programming language set: Based on big data technology, collect the source codes of executable programs written in programming languages such as C / C++ language, C#, Python, etc.

[0022] 3. Assembly instruction and machine code set: Disassemble the executable program and collect the assembly instruction set and machine code set inside the program.

[0023] 4. Establish a hierarchical multi-dimensional data set: Organize each data set into a hierarchical multi-dimensional data set of "natural language - programming language - assembly instruction - machine code". Among them, the languages between different levels show a certain topological structure. For example, the same natural language can correspond to different programming languages and assembly instruction combinations, and the same programming language can also correspond to different natural languages and assembly instruction combinations. However, there is a one-to-one correspondence between assembly instructions and machine codes, and there can be a multi-element correspondence between assembly instruction combinations and machine code combinations.

[0024] 4. Construct mapping functions: Based on statistical models such as natural language processing models, machine learning models, generalized linear models, and simple linear models, construct mapping functions between languages layer by layer until the construction of the overall mapping function system of "natural language - programming language - assembly instruction - machine code" is completed. Among them, a natural language processing model is used to construct the mapping function between natural language and programming language, a generalized linear model is used to construct the mapping function between programming language and assembly instruction, and machine learning models and simple linear models are used to construct the mapping function between assembly instruction and machine code.

[0025] 5. Generate machine code according to natural language: Send a natural language instruction to the established mapping system to generate the corresponding machine code. For example, when inputting "Print Hello World!" to the mapping system, the mapping system outputs the machine code as shown in the figure.

[0026] 6. Package the machine code into an executable program: Write an automated program to realize the batch packaging of machine code into executable files. Write an automated packaging program that can batch generate EXE executable programs from binary files through a compiler (such as Visual Studio, etc.).

[0027] 7. Test the executable program and score the intelligent system: Run and test the generated executable file, select the executable program with normal operation and less memory occupancy, and record the machine code generation scheme as a high-score scheme in the intelligent system. Here, it is necessary to perform path analysis on the topological structure between languages at each level, and by evaluating the running performance and memory occupancy of the programs generated by each feasible path, find and record the corresponding mapping path that can generate the optimal program.

[0028] 8. Iterate through processes 5 to 7 to generate a stable intelligent self-programming system: Iteratively loop the intelligent system programming process until it stably generates the optimal program. Through multiple loop iterations and tests, the system establishes the mapping path corresponding to each natural language instruction and its scoring situation. Then, machine code is generated according to the high-score mapping path function and packaged into an executable program.

Claims

1. A self-programming artificial intelligence that generates machine code based on machine learning and statistical algorithms, characterized in that, It includes the following steps: a. Establish a hierarchical multi-dimensional dataset: Organize each dataset into a hierarchical multi-dimensional dataset of "natural language - programming language - assembly instruction - machine code". Among them, the languages between different levels present a certain topological structure. For example, the same natural language can correspond to different programming languages and assembly instruction combinations, and the same programming language can also correspond to different natural languages and assembly instruction combinations. However, there is a one-to-one correspondence between assembly instructions and machine code, while there can be a multi-element correspondence between assembly instruction combinations and machine code combinations. b. Construct mapping functions: Based on statistical models such as natural language processing models, machine learning models, generalized linear models, and simple linear models, construct mapping functions between languages layer by layer until the construction of the overall mapping function system of "natural language - programming language - assembly instruction - machine code" is completed. Among them, a natural language processing model is used to construct the mapping function between natural language and programming language, a generalized linear model is used to construct the mapping function between programming language and assembly instruction, and machine learning models and simple linear models are used to construct the mapping function between assembly instruction and machine code. c. Generate machine code according to natural language: Send natural language instructions to the established mapping system to generate corresponding machine code.

2. The self-programming artificial intelligence according to claim 1, which generates machine code based on machine learning and statistical algorithms, characterized in that, The languages between different levels present a certain topological structure. For example, the same natural language can correspond to different programming languages and assembly instruction combinations, and the same programming language can also correspond to different natural languages and assembly instruction combinations. However, there is a one-to-one correspondence between assembly instructions and machine code, while there can be a multi-element correspondence between assembly instruction combinations and machine code combinations.

3. The self-programming artificial intelligence according to claim 1, which generates machine code based on machine learning and statistical algorithms, characterized in that, A natural language processing model is used to construct the mapping function between natural language and programming language, a generalized linear model is used to construct the mapping function between programming language and assembly instruction, and machine learning models and simple linear models are used to construct the mapping function between assembly instruction and machine code.

4. The self-programming artificial intelligence according to claim 1, which generates machine code based on machine learning and statistical algorithms, characterized in that, Send natural language instructions to the established mapping system to generate corresponding machine code.

Images