Responsabile: Mario Pasquato

Descrizione: A light curve is a time series of a relevant flux emitted by an astronomical source. Across various bands (X-ray, optical, gamma...) light curves can reveal a great deal about the nature of the sources and their emission processes. Time-domain astronomy turned to machine learning to process, classify, and otherwise characterize light curves. Light curves are physical measurements and as such they are affected by uncertainty, often expressed in terms of error bars.
However, out-of-the-box machine learning approaches often disregard this information. This may result in overconfident or even inaccurate predictions, and at any rate is a waste of the effort put into obtaining error bars in the first place. The aim of this thesis is to fix that by developing a machine learning pipeline for light curve analysis that seamlessly integrates error bars (and possibly otheruncertainty information such as upper or lower limits) into the analysis. We will explore solutions at the data pre-processing stage, such as treating uncertainty as an additional channel in our time series, while also looking into implementing more recent algorithms that are built to take uncertainty into account.

Durata: 9-12 Mesi

Laurea: Magistrale

Prerequisiti: Nessuno