Database-Integrated Machine Learning for Enhanced Performance

Our study presents a novel database-integrated machine learning framework aimed at bridging the gap between database systems and machine learning, addressing data transfer costs and performance concerns. The framework introduces a machine learning management system between the database client and server, effectively storing and managing fine-grained operations such as computation graphs, derivatives, calculations, and model updates within user-defined functions in the database. These operations are encapsulated using Python, enabling precise control and ease of programming in the machine learning process. Additionally, the framework supports parallel computing and performance optimization, offering outstanding performance comparable to dedicated machine learning frameworks. Most importantly, it provides a programming experience akin to traditional machine learning frameworks, allowing developers to disregard the database’s presence and focus on machine learning tasks. This research introduces a promising solution in the field of in-database machine learning, with the potential for far-reaching impacts in data-driven applications.