Highlights from two sessions at the RSS conference 2016 …

Earlier this week I popped up to the 2016 RSS Conference in Manchester to give a talk as part of the Geostatistical Models for Tropical Medicine session organised by Michelle Stanton (and featuring two other speakers: Victor Alegana from Southampton and Emanuele Giogi from Lancaster).  Given the rather steep conference fees (!) I decided to only register for one day, but nevertheless from the few talks I saw there were a couple of obvious relevance to astronomical statistics.  First, Simon Preston described the ‘ESAG’ (Elliptically Symmetric Angular Gaussian) distribution on the sphere which is constructed by projection/marginalisation of a three-dimensional Gaussian in $\mathcal{R}^3$ to the space $\mathcal{S}^2$.  Two additional conditions on the mean, $\mu$, and covariance matrix, $\Sigma$, of the three-dimensional Gaussian complete the definition of the ESAG and reduce the size of its parameter space to 5: namely, $\Sigma\mu=\mu$ and $|\Sigma|=1$.  One could well imagine using a mixture model in which the ESAG is the component distribution to represent something like the distribution of gamma-ray bursts on the sky.  Second, Timothy Cannings described the methodology behind his R package, RPEnsemble, for learning binary classifications via an ensemble of random projections from the input space, $\mathcal{R}^n$, to a lower-dimensional space, $\mathcal{R}^p$.  Given the prevalence of classification tasks in astronomical data analysis (e.g. distinguishing quasars from other bright sources in a wide-field survey catalogue) I would expect this one also to be a neat addition to the astronomers’ toolkit.