Classification by Retrieval: Binarizing Data and Classifiers

被引:36
作者
Shen, Fumin [1 ,2 ]
Mu, Yadong [3 ]
Yang, Yang [1 ,2 ]
Liu, Wei [4 ]
Liu, Li [5 ]
Song, Jingkuan [1 ,2 ]
Shen, Heng Tao [1 ,2 ]
机构
[1] Univ Elect Sci & Technol China, Ctr Future Media, Chengdu, Sichuan, Peoples R China
[2] Univ Elect Sci & Technol China, Sch Comp Sci & Engn, Chengdu, Sichuan, Peoples R China
[3] Peking Univ, Inst Comp Sci & Technol, Beijing, Peoples R China
[4] Tencent Al Lab, Bellevue, WA USA
[5] Malong Technol Co Ltd, Shenzhen, Peoples R China
来源
SIGIR'17: PROCEEDINGS OF THE 40TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL | 2017年
基金
中国国家自然科学基金;
关键词
Hashing; binary codes; classification; retrieval;
D O I
10.1145/3077136.3080767
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
This paper proposes a generic formulation that significantly expedites the training and deployment of image classification models, particularly under the scenarios of many image categories and high feature dimensions. As the core idea, our method represents both the images and learned classifiers using binary hash codes, which are simultaneously learned from the training data. Classifying an image thereby reduces to retrieving its nearest class codes in the Hamming space. Specifically, we formulate multiclass image classification as an optimization problem over binary variables. The optimization alternatingly proceeds over the binary classifiers and image hash codes. Profiting from the special property of binary codes, we show that the sub-problems can be efficiently solved through either a binary quadratic program (BQP) or a linear program. In particular, for attacking the BQP problem, we propose a novel bit-flipping procedure which enjoys high efficacy and a local optimality guarantee. Our formulation supports a large family of empirical loss functions and is, in specific, instantiated by exponential and linear losses. Comprehensive evaluations are conducted on several representative image benchmarks. The experiments consistently exhibit reduced computational and memory complexities of model training and deployment, without sacrificing classification accuracy.
引用
收藏
页码:595 / 604
页数:10
相关论文
共 54 条
[1]  
[Anonymous], 2017, P WWW
[2]  
[Anonymous], P NIPS
[3]  
[Anonymous], 2013, Advances in Neural Information Processing Systems (NIPS)
[4]  
[Anonymous], P SIGIR
[5]  
[Anonymous], 2009, NIPS
[6]  
[Anonymous], P CVPR
[7]  
[Anonymous], P ECCV
[8]  
[Anonymous], P CVPR
[9]  
[Anonymous], P CVPR
[10]   Proximal alternating linearized minimization for nonconvex and nonsmooth problems [J].
Bolte, Jerome ;
Sabach, Shoham ;
Teboulle, Marc .
MATHEMATICAL PROGRAMMING, 2014, 146 (1-2) :459-494