Latent gaussian-multinomial generative model for annotated data

Shuoran Jiang; Yarui Chen; Zhifei Qin; Jucheng Yang; Tingting Zhao; Chuanlei Zhang

doi:10.1007/978-3-030-16148-4_4

Latent gaussian-multinomial generative model for annotated data

Shuoran Jiang
, Yarui Chen^*
, Zhifei Qin
, Jucheng Yang
, Tingting Zhao
, Chuanlei Zhang

^*Corresponding author for this work

Tianjin University of Science & Technology

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Traditional generative models annotate images by multiple instances independently segmented, but these models have been becoming prohibitively expensive and time-consuming along with the growth of Internet data. Focusing on the annotated data, we propose a latent Gaussian-Multinomial generative model (LGMG), which generates the image-annotations using a multimodal probabilistic models. Specifically, we use a continuous latent variable with prior of Normal distribution as the latent representation summarizing the high-level semantics of images, and a discrete latent variable with prior of Multinomial distribution as the topics indicator for annotation. We compute the variational posteriors from a mapping structure among latent representation, topics indicator and image-annotation. The stochastic gradient variational Bayes estimator on variational objective is realized by combining the reparameterization trick and Monte Carlo estimator. Finally, we demonstrate the performance of LGMG on LabelMe in terms of held-out likelihood, automatic image annotation with the state-of-the-art models.

Original language	English
Title of host publication	Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings
Editors	Zhiguo Gong, Qiang Yang, Min-Ling Zhang, Zhi-Hua Zhou, Sheng-Jun Huang
Publisher	Springer Verlag
Pages	42-54
Number of pages	13
ISBN (Print)	9783030161477
DOIs	https://doi.org/10.1007/978-3-030-16148-4_4
State	Published - 2019
Externally published	Yes
Event	23rd Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2019 - Macau, China Duration: 14 Apr 2019 → 17 Apr 2019

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11439 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	23rd Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2019
Country/Territory	China
City	Macau
Period	14/04/19 → 17/04/19

Keywords

Annotated data
Gaussian-Multinomial
Latent representation
Multimodal generative models
Topics indicator

Access to Document

10.1007/978-3-030-16148-4_4

Cite this

Jiang, S., Chen, Y., Qin, Z., Yang, J., Zhao, T., & Zhang, C. (2019). Latent gaussian-multinomial generative model for annotated data. In Z. Gong, Q. Yang, M.-L. Zhang, Z.-H. Zhou, & S.-J. Huang (Eds.), Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings (pp. 42-54). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11439 LNAI). Springer Verlag. https://doi.org/10.1007/978-3-030-16148-4_4

Jiang, Shuoran ; Chen, Yarui ; Qin, Zhifei et al. / Latent gaussian-multinomial generative model for annotated data. Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings. editor / Zhiguo Gong ; Qiang Yang ; Min-Ling Zhang ; Zhi-Hua Zhou ; Sheng-Jun Huang. Springer Verlag, 2019. pp. 42-54 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{da12980ca71e4172a5be198f5207b33e,

title = "Latent gaussian-multinomial generative model for annotated data",

abstract = "Traditional generative models annotate images by multiple instances independently segmented, but these models have been becoming prohibitively expensive and time-consuming along with the growth of Internet data. Focusing on the annotated data, we propose a latent Gaussian-Multinomial generative model (LGMG), which generates the image-annotations using a multimodal probabilistic models. Specifically, we use a continuous latent variable with prior of Normal distribution as the latent representation summarizing the high-level semantics of images, and a discrete latent variable with prior of Multinomial distribution as the topics indicator for annotation. We compute the variational posteriors from a mapping structure among latent representation, topics indicator and image-annotation. The stochastic gradient variational Bayes estimator on variational objective is realized by combining the reparameterization trick and Monte Carlo estimator. Finally, we demonstrate the performance of LGMG on LabelMe in terms of held-out likelihood, automatic image annotation with the state-of-the-art models.",

keywords = "Annotated data, Gaussian-Multinomial, Latent representation, Multimodal generative models, Topics indicator",

author = "Shuoran Jiang and Yarui Chen and Zhifei Qin and Jucheng Yang and Tingting Zhao and Chuanlei Zhang",

note = "Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2019.; 23rd Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2019 ; Conference date: 14-04-2019 Through 17-04-2019",

year = "2019",

doi = "10.1007/978-3-030-16148-4\_4",

language = "英语",

isbn = "9783030161477",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "42--54",

editor = "Zhiguo Gong and Qiang Yang and Min-Ling Zhang and Zhi-Hua Zhou and Sheng-Jun Huang",

booktitle = "Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings",

address = "德国",

}

Jiang, S, Chen, Y, Qin, Z, Yang, J, Zhao, T & Zhang, C 2019, Latent gaussian-multinomial generative model for annotated data. in Z Gong, Q Yang, M-L Zhang, Z-H Zhou & S-J Huang (eds), Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11439 LNAI, Springer Verlag, pp. 42-54, 23rd Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2019, Macau, China, 14/04/19. https://doi.org/10.1007/978-3-030-16148-4_4

Latent gaussian-multinomial generative model for annotated data. / Jiang, Shuoran; Chen, Yarui; Qin, Zhifei et al.
Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings. ed. / Zhiguo Gong; Qiang Yang; Min-Ling Zhang; Zhi-Hua Zhou; Sheng-Jun Huang. Springer Verlag, 2019. p. 42-54 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11439 LNAI).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Latent gaussian-multinomial generative model for annotated data

AU - Jiang, Shuoran

AU - Chen, Yarui

AU - Qin, Zhifei

AU - Yang, Jucheng

AU - Zhao, Tingting

AU - Zhang, Chuanlei

N1 - Publisher Copyright: © Springer Nature Switzerland AG 2019.

PY - 2019

Y1 - 2019

N2 - Traditional generative models annotate images by multiple instances independently segmented, but these models have been becoming prohibitively expensive and time-consuming along with the growth of Internet data. Focusing on the annotated data, we propose a latent Gaussian-Multinomial generative model (LGMG), which generates the image-annotations using a multimodal probabilistic models. Specifically, we use a continuous latent variable with prior of Normal distribution as the latent representation summarizing the high-level semantics of images, and a discrete latent variable with prior of Multinomial distribution as the topics indicator for annotation. We compute the variational posteriors from a mapping structure among latent representation, topics indicator and image-annotation. The stochastic gradient variational Bayes estimator on variational objective is realized by combining the reparameterization trick and Monte Carlo estimator. Finally, we demonstrate the performance of LGMG on LabelMe in terms of held-out likelihood, automatic image annotation with the state-of-the-art models.

AB - Traditional generative models annotate images by multiple instances independently segmented, but these models have been becoming prohibitively expensive and time-consuming along with the growth of Internet data. Focusing on the annotated data, we propose a latent Gaussian-Multinomial generative model (LGMG), which generates the image-annotations using a multimodal probabilistic models. Specifically, we use a continuous latent variable with prior of Normal distribution as the latent representation summarizing the high-level semantics of images, and a discrete latent variable with prior of Multinomial distribution as the topics indicator for annotation. We compute the variational posteriors from a mapping structure among latent representation, topics indicator and image-annotation. The stochastic gradient variational Bayes estimator on variational objective is realized by combining the reparameterization trick and Monte Carlo estimator. Finally, we demonstrate the performance of LGMG on LabelMe in terms of held-out likelihood, automatic image annotation with the state-of-the-art models.

KW - Annotated data

KW - Gaussian-Multinomial

KW - Latent representation

KW - Multimodal generative models

KW - Topics indicator

UR - https://www.scopus.com/pages/publications/85064890816

U2 - 10.1007/978-3-030-16148-4_4

DO - 10.1007/978-3-030-16148-4_4

M3 - 会议稿件

AN - SCOPUS:85064890816

SN - 9783030161477

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 42

EP - 54

BT - Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings

A2 - Gong, Zhiguo

A2 - Yang, Qiang

A2 - Zhang, Min-Ling

A2 - Zhou, Zhi-Hua

A2 - Huang, Sheng-Jun

PB - Springer Verlag

T2 - 23rd Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2019

Y2 - 14 April 2019 through 17 April 2019

ER -

Jiang S, Chen Y, Qin Z, Yang J, Zhao T, Zhang C. Latent gaussian-multinomial generative model for annotated data. In Gong Z, Yang Q, Zhang ML, Zhou ZH, Huang SJ, editors, Advances in Knowledge Discovery and Data Mining - 23rd Pacific-Asia Conference, PAKDD 2019, Proceedings. Springer Verlag. 2019. p. 42-54. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-16148-4_4

Latent gaussian-multinomial generative model for annotated data

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this