The sky that a radio telescope such as the SKA observes (the feature map) is the true sky involved with a convolutional filter (the weighted sampling function). This convolution filter varies across the feature map and is different at each pixel. To recover the true sky from the observed sky or feature map, one has to deconvolve the feature map by removing all these convolutional filters at each pixel. The cost associated with this process is very large. I will be presenting a mathematical framework that is based on approximation theory to compute all these convolutional filters in the feature map with fewer computational requirements.
Dr. Marcel Atemkeng is a researcher with more than 5 years’ experience in high-bandwidth signal processing and data science with a strong background in applied mathematics, computer science, artificial intelligence, and radio interferometric data processing. He is currently developing novel dimensional reduction algorithms and applying deep learning techniques for the SKA big data. Marcel is also a lecturer in machine learning/deep learning and the coordinator of the AI research group (AIRG) at Rhodes University where he is supervising a number of students. Visit the below link to learn more about AIRG, ongoing works, and recent publications. https://www.ru.ac.za/mathematics/research/artificialintelligenceresearchgroupairg/