Efficient Hashing Methods for Finding Objects in Large Volumes of Data

Abstract

This article analyzes machine learning methods for efficient access and rapid retrieval of objects such as images, videos, and documents in large data sets. Various hashing methods are considered: deep lifelong cross-modal hashing, learned locality-sensitive hashing, graph-collaborated auto-encoder hashing for multi-view binary clustering, sparsity-induced generative adversarial hashing, contrastive language-image pre-training multimodal hashing, locality-sensitive hashing with query-based dynamic bucketing, deep supervised hashing, and cross-modal hashing methods. A computational experiment was conducted to comparatively analyze and evaluate the accuracy, loss function, and performance of the hashing algorithms: deep hashing, deep supervised hashing, and deep learning hashing. Python programs were developed for calculating the hashing algorithms and presenting graphical results. For multimodal tasks, where data from various sources must be integrated and supplemented with new data, deep lifelong cross-modal hashing is the most suitable solution. An analysis of deep hashing methods has demonstrated the superiority of deep supervised hashing when used with labeled data and distinct object classes.