Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping

Shahzad Mumtaz, Michel F. Randrianandrasana, Gurjinder Bassi, Ian T. Nabney

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Most machine-learning algorithms are designed for datasets with features of a single type whereas very little attention has been given to datasets with mixed-type features. We recently proposed a model to handle mixed types with a probabilistic latent variable formalism. This proposed model describes the data by type-specific distributions that are conditionally independent given the latent space and is called generalised generative topographic mapping (GGTM). It has often been observed that visualisations of high-dimensional datasets can be poor in the presence of noisy features. In this paper we therefore propose to extend the GGTM to estimate feature saliency values (GGTMFS) as an integrated part of the parameter learning process with an expectation-maximisation (EM) algorithm. The efficacy of the proposed GGTMFS model is demonstrated both for synthetic and real datasets.
Original languageEnglish
Title of host publicationWorkshop new challenges in neural computation 2015
EditorsBarbara Hammer, Thomas Martinetz, Thomas Villmann
Place of PublicationBielefeld (DE)
PublisherUniversität Bielefeld
Pages114-121
Number of pages8
Publication statusPublished - 1 Oct 2015
EventWorkshop new challenges in neural computation 2015 - Informatikzentrum of RWTH Aachen, Aachen, Germany
Duration: 10 Oct 201510 Oct 2015

Publication series

NameMachine learning reports
PublisherUniversität Bielefeld
Number03/2015
ISSN (Print)1865-3960

Workshop

WorkshopWorkshop new challenges in neural computation 2015
CountryGermany
CityAachen
Period10/10/1510/10/15

Fingerprint

Visualization
Learning algorithms
Learning systems

Bibliographical note

© the authors

Cite this

Mumtaz, S., Randrianandrasana, M. F., Bassi, G., & Nabney, I. T. (2015). Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping. In B. Hammer, T. Martinetz, & T. Villmann (Eds.), Workshop new challenges in neural computation 2015 (pp. 114-121). (Machine learning reports; No. 03/2015). Bielefeld (DE): Universität Bielefeld.
Mumtaz, Shahzad ; Randrianandrasana, Michel F. ; Bassi, Gurjinder ; Nabney, Ian T. / Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping. Workshop new challenges in neural computation 2015. editor / Barbara Hammer ; Thomas Martinetz ; Thomas Villmann. Bielefeld (DE) : Universität Bielefeld, 2015. pp. 114-121 (Machine learning reports; 03/2015).
@inproceedings{67d70c101d3847ceb1ed64233fa23d71,
title = "Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping",
abstract = "Most machine-learning algorithms are designed for datasets with features of a single type whereas very little attention has been given to datasets with mixed-type features. We recently proposed a model to handle mixed types with a probabilistic latent variable formalism. This proposed model describes the data by type-specific distributions that are conditionally independent given the latent space and is called generalised generative topographic mapping (GGTM). It has often been observed that visualisations of high-dimensional datasets can be poor in the presence of noisy features. In this paper we therefore propose to extend the GGTM to estimate feature saliency values (GGTMFS) as an integrated part of the parameter learning process with an expectation-maximisation (EM) algorithm. The efficacy of the proposed GGTMFS model is demonstrated both for synthetic and real datasets.",
author = "Shahzad Mumtaz and Randrianandrasana, {Michel F.} and Gurjinder Bassi and Nabney, {Ian T.}",
note = "{\circledC} the authors",
year = "2015",
month = "10",
day = "1",
language = "English",
series = "Machine learning reports",
publisher = "Universit{\"a}t Bielefeld",
number = "03/2015",
pages = "114--121",
editor = "Barbara Hammer and Thomas Martinetz and Thomas Villmann",
booktitle = "Workshop new challenges in neural computation 2015",

}

Mumtaz, S, Randrianandrasana, MF, Bassi, G & Nabney, IT 2015, Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping. in B Hammer, T Martinetz & T Villmann (eds), Workshop new challenges in neural computation 2015. Machine learning reports, no. 03/2015, Universität Bielefeld, Bielefeld (DE), pp. 114-121, Workshop new challenges in neural computation 2015, Aachen, Germany, 10/10/15.

Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping. / Mumtaz, Shahzad; Randrianandrasana, Michel F.; Bassi, Gurjinder; Nabney, Ian T.

Workshop new challenges in neural computation 2015. ed. / Barbara Hammer; Thomas Martinetz; Thomas Villmann. Bielefeld (DE) : Universität Bielefeld, 2015. p. 114-121 (Machine learning reports; No. 03/2015).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping

AU - Mumtaz, Shahzad

AU - Randrianandrasana, Michel F.

AU - Bassi, Gurjinder

AU - Nabney, Ian T.

N1 - © the authors

PY - 2015/10/1

Y1 - 2015/10/1

N2 - Most machine-learning algorithms are designed for datasets with features of a single type whereas very little attention has been given to datasets with mixed-type features. We recently proposed a model to handle mixed types with a probabilistic latent variable formalism. This proposed model describes the data by type-specific distributions that are conditionally independent given the latent space and is called generalised generative topographic mapping (GGTM). It has often been observed that visualisations of high-dimensional datasets can be poor in the presence of noisy features. In this paper we therefore propose to extend the GGTM to estimate feature saliency values (GGTMFS) as an integrated part of the parameter learning process with an expectation-maximisation (EM) algorithm. The efficacy of the proposed GGTMFS model is demonstrated both for synthetic and real datasets.

AB - Most machine-learning algorithms are designed for datasets with features of a single type whereas very little attention has been given to datasets with mixed-type features. We recently proposed a model to handle mixed types with a probabilistic latent variable formalism. This proposed model describes the data by type-specific distributions that are conditionally independent given the latent space and is called generalised generative topographic mapping (GGTM). It has often been observed that visualisations of high-dimensional datasets can be poor in the presence of noisy features. In this paper we therefore propose to extend the GGTM to estimate feature saliency values (GGTMFS) as an integrated part of the parameter learning process with an expectation-maximisation (EM) algorithm. The efficacy of the proposed GGTMFS model is demonstrated both for synthetic and real datasets.

M3 - Conference contribution

T3 - Machine learning reports

SP - 114

EP - 121

BT - Workshop new challenges in neural computation 2015

A2 - Hammer, Barbara

A2 - Martinetz, Thomas

A2 - Villmann, Thomas

PB - Universität Bielefeld

CY - Bielefeld (DE)

ER -

Mumtaz S, Randrianandrasana MF, Bassi G, Nabney IT. Visualisation of heterogeneous data with simultaneous feature saliency using Generalised Generative Topographic Mapping. In Hammer B, Martinetz T, Villmann T, editors, Workshop new challenges in neural computation 2015. Bielefeld (DE): Universität Bielefeld. 2015. p. 114-121. (Machine learning reports; 03/2015).