Exploring learned indexes for approximate query processing and visual interfaces

Abstract

Learned index structures are a promising new direction for improving data access. They offer the ability to do fast lookups in very large data sets, such as the kind needed for visual interfaces, without taking up huge amounts of memory. This paper discusses the extension of research with learned index structures as applied to approximating range search queries for visualization, some of the unexpected theoretical challenges this task brings up, and how learned index structures compare with other modern techniques for fast visualization.