Connecting Compression Spaces with Transformer for Approximate Nearest Neighbor Search

被引:2
作者
Zhang, Haokui [1 ,2 ]
Tang, Buzhou [2 ]
Hu, Wenze [1 ]
Wang, Xiaoyu [1 ]
机构
[1] Intellifusion, Shenzhen, Peoples R China
[2] Harbin Inst Technol Shenzhen, Shenzhen, Peoples R China
来源
COMPUTER VISION - ECCV 2022, PT XIV | 2022年 / 13674卷
关键词
Approximate nearest neighbor search; Transformer; Neighborhood relationship preserving; Compression projections retrieval; QUANTIZATION; ALGORITHMS;
D O I
10.1007/978-3-031-19781-9_30
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We propose a generic feature compression method for Approximate Nearest Neighbor Search (ANNS) problems, which speeds up existing ANNS methods in a plug-and-play manner. Specifically, based on transformer, we propose a new network structure to compress the feature into a low dimensional space, and an inhomogeneous neighborhood relationship preserving (INRP) loss that aims to maintain high search accuracy. Specifically, we use multiple compression projections to cast the feature into many low dimensional spaces, and then use transformer to globally optimize these projections such that the features are well compressed following the guidance from our loss function. The loss function is designed to assign high weights on point pairs that are close in original feature space, and keep their distances in projected space. Keeping these distances helps maintain the eventual top-k retrieval accuracy, and down weighting others creates room for feature compression. In experiments, we run our compression method on public datasets, and use the compressed features in graph based, product quantization and scalar quantization based ANNS solutions. Experimental results show that our compression method can significantly improve the efficiency of these methods while preserves or even improves search accuracy, suggesting its broad potential impact on real world applications.
引用
收藏
页码:515 / 530
页数:16
相关论文
共 29 条
[1]   Database-friendly random projections: Johnson-Lindenstrauss with binary coins [J].
Achlioptas, D .
JOURNAL OF COMPUTER AND SYSTEM SCIENCES, 2003, 66 (04) :671-687
[2]   Optimal Data-Dependent Hashing for Approximate Near Neighbors [J].
Andoni, Alexandr ;
Razenshteyn, Ilya .
STOC'15: PROCEEDINGS OF THE 2015 ACM SYMPOSIUM ON THEORY OF COMPUTING, 2015, :793-801
[3]  
Chen GY, 2019, Arxiv, DOI arXiv:1905.05928
[4]  
Dosovitskiy A, 2021, Arxiv, DOI [arXiv:2010.11929, DOI 10.48550/ARXIV.2010.11929]
[5]   Link and code: Fast indexing with graphs and compact regression codes [J].
Douze, Matthijs ;
Sablayrolles, Alexandre ;
Jegou, Herve .
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2018, :3646-3654
[6]  
Fu C, 2018, Arxiv, DOI [arXiv:1707.00143, 10.48550/arXiv.1707.00143]
[7]   Iterative Quantization: A Procrustean Approach to Learning Binary Codes for Large-Scale Image Retrieval [J].
Gong, Yunchao ;
Lazebnik, Svetlana ;
Gordo, Albert ;
Perronnin, Florent .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2013, 35 (12) :2916-2929
[8]  
Graham B, 2021, Arxiv, DOI [arXiv:2104.01136, 10.48550/arXiv.2104.01136]
[9]  
Guo RQ, 2020, PR MACH LEARN RES, V119
[10]   Reducing the dimensionality of data with neural networks [J].
Hinton, G. E. ;
Salakhutdinov, R. R. .
SCIENCE, 2006, 313 (5786) :504-507