Unsupervised Hashing

Unsupervised hashing aims to efficiently represent high-dimensional data using compact binary codes, preserving data similarity while reducing storage and computational costs. Recent research focuses on improving the accuracy of similarity preservation by addressing issues like "similarity collapse" through techniques such as similarity distribution calibration and incorporating richer data representations, including multi-view data and hierarchical semantic structures using autoencoders and contrastive learning within various model architectures. These advancements enhance the performance of large-scale similarity search and clustering tasks, impacting fields like image retrieval and information retrieval.