RSS

Miguel Carreira: Latent Variable Models for Nonlinear Dimensionality Reduction

Investigator: Miguel A. Carreira-Perpinan Supervisor: Steve Renals

Overview

The aim of this research project is:

  • To develop nonlinear methods that reduce the dimensionality of high-dimensional vector streams (e.g. images, auditory model data) but preserving the fine detail information of the original data. In particular, continuous latent variable models will be considered. This broad class of probabilistic models encompasses, among others, factor analysis, principal component analysis (PCA), the generative topographic mapping (GTM), independent component analysis (ICA) and mixture distributions.
  • To apply the results to electropalatography (EPG) data and to other speech data and compare them with the results of linear methods (e.g. principal component analysis and factor analysis).
  • To provide with low-dimensional representations of the electropalatography (EPG) data, more amenable to analysis and with applications to speech therapy, language learning and articulatory dynamics.
  • To approach the problem of the acoustic-to-articulatory mapping problem by constructing functional relationships from the conditional distributions of the latent variable model.

This research has been supported by the Spanish Ministry of Education and Science and by the SPRACH project.

Please see my research web page for a more detailed explanation of this research.

References

  • Carreira-Perpinan, M. A. and Renals, S. (1998): "A latent variable modelling approach to the acoustic-to-articulatory mapping problem". To appear in the Proc. of the 14th International Congress of Phonetic Sciences (ICPhS'99), San Francisco, 1-7 August 1999.
  • Carreira-Perpinan, M. A. and Renals, S. (1998): Dimensionality reduction of electropalatographic data using latent variable models. Speech Communication 26(4), pp. 259-282. Online version.
  • Carreira-Perpinan, M. A. and Renals, S. (1998): "Experimental evaluation of latent variable models for dimensionality reduction". Proc. of the 1998 IEEE Signal Processing Society Workshop on Neural Networks for Signal Processing (NNSP98), pp. 165-173, Cambridge, UK. Online version.
  • Carreira-Perpinan, M. A. and Renals, S. (1998): Practical identifiability of finite mixtures of multivariate Bernoulli distributions. Submitted. Online version.
  • Carreira-Perpinan, M. A. (1997): Dimension reduction of articulatory data. Poster presented at the NATO ASI Generalization in Neural Networks and Machine Learning, Isaac Newton Institute, Cambridge, 4-15 Aug. 1997. Online version.
  • Carreira-Perpinan, M. A. (1997): Density networks for dimension reduction of continuous data: Analytical solutions. Technical report CS-97-09, Dept. of Computer Science, University of Sheffield, UK. Online version.
  • Carreira-Perpinan, M. A. (1996): A review of dimension reduction techniques. Technical report CS-96-09, Dept. of Computer Science, University of Sheffield, UK. Online version.

Please see my papers page for online copies of my papers and associated software.