Learning from Data to Speed-up Sorted Table Search Procedures: Methodology and Practical Guidelines

TL;DR

This paper explores how machine learning can accelerate sorted table search procedures, providing a systematic comparison of traditional and learned methods across CPU and GPU, and formalizing a new paradigm for learned search algorithms.

Contribution

It introduces a formalized paradigm for learned dichotomic search procedures and compares their performance with traditional methods across different data layouts and hardware.

Findings

Learned search procedures can outperform traditional methods in specific scenarios.

The effectiveness of learned methods depends on data layout and hardware used.

A new formal framework for learned sorted table search algorithms is proposed.

Abstract

Sorted Table Search Procedures are the quintessential query-answering tool, with widespread usage that now includes also Web Applications, e.g, Search Engines (Google Chrome) and ad Bidding Systems (AppNexus). Speeding them up, at very little cost in space, is still a quite significant achievement. Here we study to what extend Machine Learning Techniques can contribute to obtain such a speed-up via a systematic experimental comparison of known efficient implementations of Sorted Table Search procedures, with different Data Layouts, and their Learned counterparts developed here. We characterize the scenarios in which those latter can be profitably used with respect to the former, accounting for both CPU and GPU computing. Our approach contributes also to the study of Learned Data Structures, a recent proposal to improve the time/space performance of fundamental Data Structures, e.g., B-trees, Hash Tables, Bloom Filters. Indeed, we also formalize an Algorithmic Paradigm of Learned Dichotomic Sorted Table Search procedures that naturally complements the Learned one proposed here and that characterizes most of the known Sorted Table Search Procedures as having a "learning phase" that approximates Simple Linear Regression.