Automated Machine Learning Research On PAR Platform

Software automation refers to the process of generating software automatically based on formal or informal specifications.This has been a long-standing dream in the field of computer science,liberating developers from the burden of monotonous programming tasks.Algorithms,known as the "soul" of computing,are the backbone of software,and their design plays a crucial role in determining the reliability of software systems.Currently,although some studies have attempted to utilize machine learning algorithms such as deep learning to achieve algorithm design automation,the creative and uncertain nature of the algorithm design process,the large differences in algorithm program expression,the issues with algorithm development data quality,and the wide range of problems involved in algorithm design all contribute to the limited level of automation achieved thus far.In light of this,our research builds upon Professor Xue Jinyun’s PAR method and platform to conduct an accurate vectorization of the Radl specification,and proposes a vectorization transformation method.With a focus on the context of general artificial intelligence,we explore the direction of autonomous intelligence generation from Radl specification to Radl algorithm,and employ the elements of machine learning automation design based on learning methodology to develop a Radl specification to Radl algorithm automatic generation system.Finally,we conduct a comparative analysis of the verifiability and reliability of traditional design methods and system generation methods.The work and innovation of this dissertation mainly focus on the following aspects:1.After studying the current mathematical expression vectorization methods,this dissertation has conducted precise vectorization research on Radl specifications,and proposed an innovative Radl specification vectorization conversion method.This method not only preserves the mathematical entity information of Radl specifications,but also includes the frequency of occurrence of mathematical entities in Radl specifications.The innovation of this method lies in solving the problem that only a two-dimensional vector transformation of mathematical formulas cannot count the frequency information of an entity that appears multiple times.2.Drawing from the theory of automatic machine learning,the objective of designing a system for automatic machine learning is posited in this study.Furthermore,the direction of autonomous intelligent generation of Radl algorithm is identified,and a pattern library for Radl algorithm is constructed.Finally,the design principle and overall framework of the system are presented.