Abstract
Indexes are essential for efficient information retrieval. This thesis presents LESIM
(LEarned Segmentation Index with Multiple pointers), a learned structure for indexing
records based on their timestamps. LESIM overcomes limitations of existing index
structures, and supports efficient updates at the current time, as well as point queries
over the past and the present. Extensive experiments were conducted based on two
common real-world datasets. In comparison to the state-of-the-art learned Piecewise
Geometric Model index (PGM), results demonstrate that LESIM provides a significant
improvement in query and append performance. However, this comes at the cost of
increased space consumption. The build time is competitive.
Post a Comment