Date: 2025-05-14
Time: 10:00–10:50
Room: Breakout Room 3
Level: Intermediate
Traditional indexing structures, such as B-trees and hash indexes, have been foundational in optimising database query performance for decades. However, recent advancements in machine learning have introduced “learned indexes,” a novel approach to data indexing that leverages statistical models to predict data locations. This method can potentially outperform traditional indexes in read-heavy workloads, especially in scenarios where data distributions are skewed or non-uniform.
In this talk, we’ll dive into the principles of learned indexes, explore how these ML-driven models aim to replace or supplement conventional indexing methods. We’ll cover the potential benefits, including lower memory footprint, reduced storage overhead, and faster lookup times for certain query types. Additionally, we’ll address the challenges and trade-offs of integrating learned indexes into PostgreSQL, where durability, adaptability, and compatibility with existing query optimisation techniques are crucial.
Attendees will come away with a comprehensive understanding of learned indexes’ potential benefits and limitations real-world PostgreSQL environments.