Machine Learning for Databases: Foundations, Paradigms, and Open problems
Gao Cong, Jingyi Yang, Yue Zhao
Abstract
This tutorial delves into the burgeoning field of Machine Learning for Databases (ML4DB), highlighting its recent progress and the challenges impeding its integration into industrial-grade database management systems. We systematically explore three key themes: the exploration of foundations in ML4DB and their potential for diverse applications, the two paradigms in ML4DB, i.e., using machine learning as a replacement versus enhancement of traditional database components, and the critical open challenges such as improving model efficiency and addressing data shifts. Through an in-depth analysis, including a survey of recent works in major database conferences, this tutorial encapsulates the current state of ML4DB, as well as charts a roadmap for its future development and wider adoption in practical database environments.