Back to Search
Start Over
Dynamic programming based optimized product quantization for approximate nearest neighbor search
- Source :
- Neurocomputing. 217:110-118
- Publication Year :
- 2016
- Publisher :
- Elsevier BV, 2016.
-
Abstract
- Product quantization and its variances have emerged in approximate nearest neighbor search, with a wide range of applications. However, the optimized division of product subspaces retains as an open problem that largely degenerates the retrieval accuracy. In the paper, an extremely optimized product quantization scheme is introduced, which ensures, both theoretically and experimentally, a much better subspace partition comparing to the existing state-of-the-arts PQ and OPQ. The key innovation is to formulate subspace partition as a graph-based optimization problem, by which dynamic programming is leveraged to pursuit optimal quantizer learning. Another advantage is that the proposed scheme is very easily integrated with the cutting-edge multi-indexing structure, with a nearly eligible overhead in addition. We have conducted a serial of large-scale quantitative evaluations, with comparisons to a group of recent works including PQ, OPQ, and multi-Index. We have shown superior performance gain in the widely used SIFT1B benchmark, which validates the merits of the proposed algorithm.
- Subjects :
- Mathematical optimization
Optimization problem
Cognitive Neuroscience
Nearest neighbor search
02 engineering and technology
010501 environmental sciences
01 natural sciences
Computer Science Applications
Dynamic programming
Artificial Intelligence
Product (mathematics)
0202 electrical engineering, electronic engineering, information engineering
Benchmark (computing)
Graph (abstract data type)
Overhead (computing)
020201 artificial intelligence & image processing
Subspace topology
0105 earth and related environmental sciences
Mathematics
Subjects
Details
- ISSN :
- 09252312
- Volume :
- 217
- Database :
- OpenAIRE
- Journal :
- Neurocomputing
- Accession number :
- edsair.doi...........42081a84fb6738f289815e8950620917
- Full Text :
- https://doi.org/10.1016/j.neucom.2016.01.112