Virtual Brown Bag Lunch and Marlar Screening
Monday, May 2, 2022 at 12:00 Link Below
Join BBL Zoom Meeting
Presentation in Marlar for those wishing to attend in person
Alex Gagliano (University of Illinois at Urbana Champaign; Flatiron Institute)
Leveraging Local Correlations to Enable Fast Transient Studies
By demanding robust processing pipelines for hundreds of thousands (and soon tens of millions) of nightly alerts, astronomical surveys continue to transform the way we study the time-domain sky. Observatories such as Roman and Rubin are slated to jointly discover millions of supernovae, many at high redshift (z~1); however, our ability to characterize each discovered event in detail has already stagnated due to a dearth of complementary spectroscopic resources for follow-up. In this talk, I will outline my work to leverage contextual information to expand our suite of tools for characterizing and prioritizing detected events. After providing a brief overview into the connection between a transient and its host galaxy, I will introduce SCOTCH, the first synthetic catalog containing millions of local and cosmological transients (0.01 < z < 3) with realistic host galaxies. SCOTCH will play a decisive role in validating transient analysis software in advance of Rubin, Roman, and Webb first light.
Alex Gagliano is a 4th year NSF Fellow at the University of Illinois at Urbana-Champaign and a Pre-Doctoral Fellow at Flatiron’s Center for Computational Astrophysics. He studies the statistical and event-specific correlations between supernovae and their environments. Alex is interested in using these relationships to better characterize the progenitor physics of explosive transients in massive surveys, particularly for rare events.
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
at 12:30 pm
Matyas Molnar (Cambridge UK)
Robust and alternative statistical methods for radio interferometry, with applications to HERA
In the era of big data in radio astronomy, large data volumes need to be reduced, much of which is corrupted by RFI, systematics and other effects. Most outlier rejection algorithms are ad-hoc and can be intricate and costly; these can miss anomalous effects as well as cause overflagging. On the other hand, robust statistical techniques can be used easily and directly on the data for location and scale estimation, and have good performance on non-normal/contaminated data owing to their high breakdown point. In this talk, I present a couple of robust statistical techniques: I introduce an outlier detection algorithm that uses MCD-based Mahalanobis distances, and the geometric median location estimator. I apply these to aggregated calibrated HERA visibilities in Local Sidereal Time, and use geometric median estimated visibilities to compute power spectra. Following on from the HERA 2022 Limits paper results, this analysis excludes the hypothesis that low-level RFI that does not repeat nightly, is corrupting the power spectrum from Band 2 Field 2. We also set a tentative new limit on the 21 cm power spectrum from these robust results. I will also discuss multiresolution analysis with wavelets as a way to better represent the 21 cm signal, which is non-stationary across redshifts due to the lightcone effect. I also show how the wavelet transform can be used for error detection by inspecting frequency-delay scaleograms.
Matyas Molnar is a final-year PhD student at the University of Cambridge and is supervised by Dr Bojan Nikolic. His doctoral research focuses on applying robust statistical techniques to the calibration and reduction of radio interferometric visibilities for 21 cm cosmology.