Weak Supervision for RFI Mitigation in 21-cm Cosmology

Speakers: Dr Shawn Dubey

Detecting the redshifted 21-cm signal from the early universe requires separating an extremely faint cosmological signal from bright foreground emission and terrestrial radio-frequency interference (RFI), but flagging RFI is a big challenge. I will describe our use of weak supervision, a method that combines multiple imperfect flaggers to construct training labels for artificial intelligence/machine learning (AI/ML) models. I will discuss the problem, standard methods, and initial AI/ML approaches. This work illustrates how AI/ML can be adapted to scientific measurements where the labels themselves are uncertain.

https://indico.phys.hawaii.edu/event/2776