Discover supernovae in the K2 dataset

[gully]

Description

K2 has observed over 40,000 galaxies anticipating ~50-100 Supernova outbursts. Instrumental artifacts arising from spacecraft motion have hindered the unambiguous census of transient signals in K2 data. In this research project, one would develop and apply data analysis methods to detect and characterize astrophysical transients observed by K2. Possible paths include difference imaging and scene modeling.

Benefits

• Potentially discover new classes of transients or previously un-identified supernovae

Costs (and anticipated difficulties)

• Distinguishing real signal from noise blips will be difficult, since correlated space-craft motion can resemble real transients.

[mirca]

Cotrending Basis Vectors (CBV) correction would possibly come in handy here (@barentsen @gully @christinahedges correct me if I'm wrong).

Right now the CBV correction happens by find the maximum a posteriori probability (MAP) estimator of Y ~ L(Σai*cbvi, s), in which Y is a light curve, ~ means ''distributed as'', L stands for the Laplacian distributions, the sum Σ is over i, ai and cbvi are the i-th CBV coefficient and CBV, respectively; and s is proportional to the measurement uncertainty. The joint prior distribution over the coefficients ai is Laplacian with zero mean and variance not much bigger than 10 (i.e., we expect the coefficients to be small and sparse).

@barentsen once suggested me that we could generalize that procedure to Y ~ L(f(θ) + Σai*cbvi, s), in which f is some functional that describes some astrophysical signal of interest and θ are the parameters of interest.

I think we could plug sncosmo models into that procedure?