Analogy-preserving Semantic Embedding for Visual Object Categorization

Sung Ju Hwang, Kristen Grauman, Fei Sha
Proceedings of the 30th International Conference on Machine Learning, PMLR 28(3):639-647, 2013.

Abstract

In multi-class categorization tasks, knowledge about the classes’ semantic relationships can provide valuable information beyond the class labels themselves. However, existing techniques focus on preserving the semantic distances between classes (e.g., according to a given object taxonomy for visual recognition), limiting the influence to pairwise structures. We propose to model \emphanalogies that reflect the relationships between multiple pairs of classes simultaneously, in the form “p is to q, as r is to s"". We translate semantic analogies into higher-order geometric constraints called \emphanalogical parallelograms, and use them in a novel convex regularizer for a discriminatively learned label embedding. Furthermore, we show how to discover analogies from attribute-based class descriptions, and how to prioritize those likely to reduce inter-class confusion. Evaluating our Analogy-preserving Semantic Embedding (ASE) on two visual recognition datasets, we demonstrate clear improvements over existing approaches, both in terms of recognition accuracy and analogy completion.

Cite this Paper


BibTeX
@InProceedings{pmlr-v28-juhwang13, title = {Analogy-preserving Semantic Embedding for Visual Object Categorization}, author = {Ju Hwang, Sung and Grauman, Kristen and Sha, Fei}, booktitle = {Proceedings of the 30th International Conference on Machine Learning}, pages = {639--647}, year = {2013}, editor = {Dasgupta, Sanjoy and McAllester, David}, volume = {28}, number = {3}, series = {Proceedings of Machine Learning Research}, address = {Atlanta, Georgia, USA}, month = {17--19 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v28/juhwang13.pdf}, url = {https://proceedings.mlr.press/v28/juhwang13.html}, abstract = {In multi-class categorization tasks, knowledge about the classes’ semantic relationships can provide valuable information beyond the class labels themselves. However, existing techniques focus on preserving the semantic distances between classes (e.g., according to a given object taxonomy for visual recognition), limiting the influence to pairwise structures. We propose to model \emphanalogies that reflect the relationships between multiple pairs of classes simultaneously, in the form “p is to q, as r is to s"". We translate semantic analogies into higher-order geometric constraints called \emphanalogical parallelograms, and use them in a novel convex regularizer for a discriminatively learned label embedding. Furthermore, we show how to discover analogies from attribute-based class descriptions, and how to prioritize those likely to reduce inter-class confusion. Evaluating our Analogy-preserving Semantic Embedding (ASE) on two visual recognition datasets, we demonstrate clear improvements over existing approaches, both in terms of recognition accuracy and analogy completion. } }
Endnote
%0 Conference Paper %T Analogy-preserving Semantic Embedding for Visual Object Categorization %A Sung Ju Hwang %A Kristen Grauman %A Fei Sha %B Proceedings of the 30th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2013 %E Sanjoy Dasgupta %E David McAllester %F pmlr-v28-juhwang13 %I PMLR %P 639--647 %U https://proceedings.mlr.press/v28/juhwang13.html %V 28 %N 3 %X In multi-class categorization tasks, knowledge about the classes’ semantic relationships can provide valuable information beyond the class labels themselves. However, existing techniques focus on preserving the semantic distances between classes (e.g., according to a given object taxonomy for visual recognition), limiting the influence to pairwise structures. We propose to model \emphanalogies that reflect the relationships between multiple pairs of classes simultaneously, in the form “p is to q, as r is to s"". We translate semantic analogies into higher-order geometric constraints called \emphanalogical parallelograms, and use them in a novel convex regularizer for a discriminatively learned label embedding. Furthermore, we show how to discover analogies from attribute-based class descriptions, and how to prioritize those likely to reduce inter-class confusion. Evaluating our Analogy-preserving Semantic Embedding (ASE) on two visual recognition datasets, we demonstrate clear improvements over existing approaches, both in terms of recognition accuracy and analogy completion.
RIS
TY - CPAPER TI - Analogy-preserving Semantic Embedding for Visual Object Categorization AU - Sung Ju Hwang AU - Kristen Grauman AU - Fei Sha BT - Proceedings of the 30th International Conference on Machine Learning DA - 2013/05/26 ED - Sanjoy Dasgupta ED - David McAllester ID - pmlr-v28-juhwang13 PB - PMLR DP - Proceedings of Machine Learning Research VL - 28 IS - 3 SP - 639 EP - 647 L1 - http://proceedings.mlr.press/v28/juhwang13.pdf UR - https://proceedings.mlr.press/v28/juhwang13.html AB - In multi-class categorization tasks, knowledge about the classes’ semantic relationships can provide valuable information beyond the class labels themselves. However, existing techniques focus on preserving the semantic distances between classes (e.g., according to a given object taxonomy for visual recognition), limiting the influence to pairwise structures. We propose to model \emphanalogies that reflect the relationships between multiple pairs of classes simultaneously, in the form “p is to q, as r is to s"". We translate semantic analogies into higher-order geometric constraints called \emphanalogical parallelograms, and use them in a novel convex regularizer for a discriminatively learned label embedding. Furthermore, we show how to discover analogies from attribute-based class descriptions, and how to prioritize those likely to reduce inter-class confusion. Evaluating our Analogy-preserving Semantic Embedding (ASE) on two visual recognition datasets, we demonstrate clear improvements over existing approaches, both in terms of recognition accuracy and analogy completion. ER -
APA
Ju Hwang, S., Grauman, K. & Sha, F.. (2013). Analogy-preserving Semantic Embedding for Visual Object Categorization. Proceedings of the 30th International Conference on Machine Learning, in Proceedings of Machine Learning Research 28(3):639-647 Available from https://proceedings.mlr.press/v28/juhwang13.html.

Related Material