A Prototype for Science Alerts

Many of the science goals of the GAIA mission, especially for bursting or time-varying phenomena like supernovae or microlensing, require an early identification, analysis and release of preliminary data. The alerting on rare and unusual events is the scientific equivalent of the finding of needles in haystacks or the panning for gold-dust in rivers. In modern signal processing, such novelty detection is routinely performed with self-organising maps (SOMs), which are an unsupervised clustering algorithm invented by Kohonen. Here, we describe the application of SOMS to the classification of data provided by large-scale surveys such as GAIA and to the despatching of scientific alerts. We illustrate our ideas by processing the publicly available OGLE II dataset towards the Bulge, identifying major classes of variable stars (such as novae, small amplitude red giant variables, eclipsing binaries and so on) and extracting the rare, discrepant lightcurves from which the alerts can be drawn