The right measure of similarity between examples is important in many areas of computer science. In particular it is a critical component in example-based learning methods. Similarity is commonly defined in terms of a conventional distance function, but such a definition does not necessarily capture the inherent meaning of similarity, which tends to depend on the underlying task. We develop an algorithmic approach to learning similarity from examples of what objects are deemed similar according to the task-specific notion of similarity at hand, as well as optional negative examples. Our learning algorithm constructs, in a greedy fashion, an encoding of the data. This encoding can be seen as an embedding into a space, where a weighted Hamming distance is correlated with the unknown similarity. This allows us to predict when two previously unseen examples are similar and, importantly, to efficiently search a very large database for examples similar to a query. This approach is tested on a set of standard machine learning benchmark problems. The model of similarity learned with our algorithm provides and improvement over standard example-based classification and regression. We also apply this framework to problems in computer vision: articulated pose estimation of humans from single images, articulated tracking in video, and matching image regions subject to generic visual similarity. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)
Cited By
- Liu J, Rahmani H, Akhtar N and Mian A (2019). Learning Human Pose Models from Synthesized Data for Robust RGB-D Action Recognition, International Journal of Computer Vision, 127:10, (1545-1564), Online publication date: 1-Oct-2019.
- Zieba M, Semberecki P, El-Gaaly T and Trzcinski T BinGAN Proceedings of the 32nd International Conference on Neural Information Processing Systems, (3612-3622)
- Yan T, Xu X, Guo S, Huang Z and Wang X Supervised Robust Discrete Multimodal Hashing for Cross-Media Retrieval Proceedings of the 25th ACM International on Conference on Information and Knowledge Management, (1271-1280)
- Wang F, Zhao W, Ngo C and Merialdo B (2014). A Hamming Embedding Kernel with Informative Bag-of-Visual Words for Video Semantic Indexing, ACM Transactions on Multimedia Computing, Communications, and Applications, 10:3, (1-20), Online publication date: 1-Apr-2014.
- Pengcheng G, Jiangqin W, Yuan L, Yang X, Tianjiao M and Baogang W Fast image-based Chinese calligraphic character retrieval on large scale data Proceedings of the 14th ACM/IEEE-CS Joint Conference on Digital Libraries, (211-218)
- Zhen Y and Yeung D A probabilistic model for multimodal hash function learning Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining, (940-948)
- Trzcinski T and Lepetit V Efficient discriminative projections for compact binary descriptors Proceedings of the 12th European conference on Computer Vision - Volume Part I, (228-242)
- Boix X, Roig G, Leistner C and Gool L Nested Sparse Quantization for Efficient Feature Coding Proceedings, Part II, of the 12th European Conference on Computer Vision --- ECCV 2012 - Volume 7573, (744-758)
- Bronstein A, Bronstein M, Guibas L and Ovsjanikov M (2011). Shape google, ACM Transactions on Graphics, 30:1, (1-20), Online publication date: 1-Jan-2011.
- Wang J, Kumar S and Chang S Sequential projection learning for hashing with compact codes Proceedings of the 27th International Conference on International Conference on Machine Learning, (1127-1134)
- Calonder M, Lepetit V, Strecha C and Fua P BRIEF Proceedings of the 11th European conference on Computer vision: Part IV, (778-792)
- Baluja S and Covell M Learning "forgiving" hash functions Proceedings of the 20th international joint conference on Artifical intelligence, (2663-2669)
Index Terms
- Learning task-specific similarity
Recommendations
Learning similarity with cosine similarity ensemble
This paper proposes a cosine similarity ensemble (CSE) method to learn similarity.CSE is a selective ensemble and combines multiple cosine similarity learners.A learner redefines the pattern vectors and determines its threshold adaptively.Experimental ...
A metric learning based approach to evaluate task-specific time series similarity
WAIM'13: Proceedings of the 14th international conference on Web-Age Information ManagementA variety of methods have been proposed to measure time series similarity, such as Dynamic Time Warping and Edit distance. Although these methods have been shown to be effective and useful in various data mining tasks, they seldom consider task-specific ...
Learning similarity measures from data
AbstractDefining similarity measures is a requirement for some machine learning methods. One such method is case-based reasoning (CBR) where the similarity measure is used to retrieve the stored case or a set of cases most similar to the query case. ...