Machine Learning-Driven Web System for Automated Sequential-to-Parallel Code Transformation

This paper presents the design and implementation of a machine learning-driven web system for automating the transformation of sequential code into parallel code. The proposed system leverages a shared memory programming model and is built on a microservices architecture to ensure scalability and modularity. It incorporates advanced tools for constructing and analyzing Abstract Syntax Trees to generate parallel programs that adhere to Bernstein's conditions. A supervised learning mechanism, using Word2Vec for vector representation and TF-IDF for significance evaluation, enables the system to adapt and refine parallelization rules based on programmer input. Additionally, the system integrates an AMQP-based message broker to handle high-load processing efficiently and supports data storage through relational or NoSQL databases. The solution provides a robust and flexible platform for enhancing computational performance and reducing development complexity in parallel programming. The results demonstrate the feasibility and effectiveness of this approach in addressing the challenges of automating code parallelization in modern software systems.