Year: 2025
Type: article
Abstract: Traditional indexing techniques, such as B-trees and hashing, have been widely used to enable efficient data retrieval in large-scale databases.However, these methods typically require relatively large index files, making it challenging to guarantee query performance in computing environments with limited main memory capacity.In contrast, Bloom filters support query processing using small-sized bit arrays and lightweight computations, making them well-suited for memory-constrained systems.However, standard Bloom filters suffer from two limitations.First, deleting index entries is problematic because removing a bit may affect the membership information of other objects, effectively preventing safe deletion.Second, both the number of elements to be indexed and the size of the bit array must be predetermined to ensure optimal performance, which poses complex challenges in dynamic environments where the dataset grows over time.To deal with these limitations, several Bloom filter variants tailored to dynamic indexing environments have been proposed.Our study aims to analyze and organize existing Bloom filter research as a reference for future development of new techniques suitable for dynamic indexing environments.Through our analysis, we concluded that meaningful performance improvements are difficult to achieve solely through modifications to the bit array structure or enhancements to deletion mechanisms.Accordingly, we propose a new direction: integrating AI-based learned indexes with Bloom filters as a hybrid model for enhanced performance.Although this model may not be the only possible solution, similar approaches could inspire the development of more efficient next-generation Bloom filter techniques.
Author: Seong Chae Lim
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4413835418