The use of machine learning in construction of the Learned Bloom Filter

The aim of this work is to construct and compare the effectiveness of the Learned Bloom Filter to a traditional Bloom filter. For this purpose, the results on three different datasets, divided into sets with categorical and text data are compared. The theoretical basis of the Bloom filter and the Learned Bloom filter designs are presented, along with their properties. The details of the design of the binary classifiers based on artificial neural networks, used in the Learned Bloom Filter, are discussed, with particular emphasis on the choice of architecture of the networks in terms of their size. The experiments were chosen to highlight problems and their solutions encountered during the design of the Learned Bloom Filter. Based on the experiments, the features of the dataset for which the use of a Learned Bloom Filter has the potential to bring benefits are briefly discussed.