Poster Abstract

P3.21 Pius Nyanumba (University of Nairobi)

Theme: Data science challenges: tools from statistics to machine learning

Machine Learning: Gaussian Process Modelling in Rotation Measure synthesis

This project is looking at Gaussian Process Modelling as a method of Machine Learning to improve transfer functions in Rotation measure synthesis. When performing Rotation Measure Synthesis, Radio Frequency Interference flagging and uneven band-pass weighting creates high non-uniform sampling patterns in spectral polarization data. This non-uniformity leads to poor transfer functions which causes problems in the interpretation of data due to the problems of the deconvolving complex spectra. This project considers Bayessian modelling for Gaussian Processes as the approach to predict the missing data affected by Radio Frequency Interfernce by calculating the relative behavior between data
points at different separations. Then calculate the posterior probability at each data point and use joint probability distribution to recover the missing data.