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

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