Comprehensive observations of variable stars can include time domain photometry in a multitude of filters, spectroscopy, estimates of color (e.g. U-B), etc. When it is considered that the time domain data can be further transformed via digital signal processing methodologies, the potential representations of the observed target star are limitless. If the goal is classification of observed variable stars, using this multitude of representations/views can become a challenge as many of the modern pattern classification algorithms in industry are limited to single input and single output. Presented here is an initial review of multi-view classification as applied to variable star classification, to address this challenge. The variable star UCR Starlight dataset and LINEAR dataset are used to generate a baseline performance estimate. The LM^3L algorithm is applied to a set of generic features, and the performance with regard to a generic feature space is evaluated. A matrix-variate implementation of the LM^3L algorithm is designed and presented specifically for this task as well; the matrix-variate implementations are novel developments.
Link to PDF (may not be available yet): P1-4.pdf