Dynamic Distal Spatial Approximation Trees

Metric space indexes are critical for efficient similarity searches across various applications. The Distal Spatial Approximation Tree (DiSAT) has demonstrated exceptional speed/memory trade-offs without requiring parameter tuning. However, since it operates solely on static databases, its application is limited in many exciting use cases. This research has been dedicated to developing a dynamic version of DiSAT that allows for incremental construction. It is remarkable that the dynamic version is faster than its static counterpart. The outcome is a faster index with the same memory requirements as DiSAT. This development enhances the practicality of DiSAT, unlocking a wide range of proximity database applications.