The Case for Learned Spatial Indexes
This work addresses the need for faster spatial data processing in applications like GPS devices and location-based services, but it is incremental as it builds on existing learned index methods.
The paper tackled the problem of efficiently processing spatial data by applying learned index techniques to classical multi-dimensional indexes for spatial range queries, resulting in speed improvements of 11.79% to 39.51% for machine learned search over binary search and 1.23x to 1.83x faster performance compared to competitors.
Spatial data is ubiquitous. Massive amounts of data are generated every day from billions of GPS-enabled devices such as cell phones, cars, sensors, and various consumer-based applications such as Uber, Tinder, location-tagged posts in Facebook, Twitter, Instagram, etc. This exponential growth in spatial data has led the research community to focus on building systems and applications that can process spatial data efficiently. In the meantime, recent research has introduced learned index structures. In this work, we use techniques proposed from a state-of-the art learned multi-dimensional index structure (namely, Flood) and apply them to five classical multi-dimensional indexes to be able to answer spatial range queries. By tuning each partitioning technique for optimal performance, we show that (i) machine learned search within a partition is faster by 11.79\% to 39.51\% than binary search when using filtering on one dimension, (ii) the bottleneck for tree structures is index lookup, which could potentially be improved by linearizing the indexed partitions (iii) filtering on one dimension and refining using machine learned indexes is 1.23x to 1.83x times faster than closest competitor which filters on two dimensions, and (iv) learned indexes can have a significant impact on the performance of low selectivity queries while being less effective under higher selectivities.