PROJECT TITLE :

Memory Vectors for Similarity Search in High-Dimensional Spaces - 2018

ABSTRACT:

We tend to study an indexing design to store and search in a very database of high-dimensional vectors from the perspective of statistical Signal Processing and call theory. This architecture consists of several memory units, each of which summarizes a fraction of the database by a single representative vector. The potential similarity of the question to 1 of the vectors stored in the memory unit is gauged by a simple correlation with the memory unit's representative vector. This representative optimizes the check of the following hypothesis: the query is freelance from any vector in the memory unit versus the query is a straightforward perturbation of one of the stored vectors. Compared to exhaustive search, our approach finds the most similar database vectors considerably faster while not a comprehensible reduction in search quality. Interestingly, the reduction of complexity is provably higher in high-dimensional spaces. We empirically demonstrate its sensible interest in a massive-scale image search state of affairs with off-the-shelf state-of-the-art descriptors.