Deep Hashing using Entropy Regularised Product Quantisation Network
This work addresses the need for flexible similarity measures in large-scale data retrieval, particularly when labels are unavailable or input variability is high, though it appears incremental as it builds on existing deep hashing paradigms.
The authors tackled the problem of approximate nearest neighbor search in large-scale systems by proposing a deep hashing method that transforms data to a uniform distribution on a product of spheres and maximizes output entropy. They demonstrated that their method outperforms baseline approaches like locality-sensitive hashing and product quantisation in limited capacity regimes.
In large scale systems, approximate nearest neighbour search is a crucial algorithm to enable efficient data retrievals. Recently, deep learning-based hashing algorithms have been proposed as a promising paradigm to enable data dependent schemes. Often their efficacy is only demonstrated on data sets with fixed, limited numbers of classes. In practical scenarios, those labels are not always available or one requires a method that can handle a higher input variability, as well as a higher granularity. To fulfil those requirements, we look at more flexible similarity measures. In this work, we present a novel, flexible, end-to-end trainable network for large-scale data hashing. Our method works by transforming the data distribution to behave as a uniform distribution on a product of spheres. The transformed data is subsequently hashed to a binary form in a way that maximises entropy of the output, (i.e. to fully utilise the available bit-rate capacity) while maintaining the correctness (i.e. close items hash to the same key in the map). We show that the method outperforms baseline approaches such as locality-sensitive hashing and product quantisation in the limited capacity regime.