Abstract

Invited Talk - Splinter DataManage

Thursday, 12 September 2024, 14:00   (S22)

Identification of pulsar signals in large data streams using machine learning and digital twins

Dr. Gautam Dange (4), Dr. Lars Haupt (5), Mr. Andrei Kazantsev (2), Dr. Yurii Pidopryhora (2), Dr. Tanumoy Saha (1), Mr. Marcel Trattner (1), Prof. Frank Bertoldi (3), Hermann Heßling (1, 5)
(1) HTW Berlin, (2) MPIfR Bonn, (3) U Bonn, (4) FIAS Frankfurt, (5) DZA Görlitz

To measure pulsar signals already during data acquisition, a novel Machine Learning-based Pipeline for Pulsar Analysis (ML-PPA) is being developed. Identifying noisy signals in real-time is a challenge, especially given the high data rates of the Square Kilometre Array Observatory (SKAO). An interdisciplinary team of astronomers and computer scientists is cooperating to realize the "software to the data" paradigm. Considerable amounts of synthetic data (digital twins) need to be generated in order to effectively train the machine learning algorithms. The path of pulsar signals from a source to radio telescopes is simulated. To optimize the scalability of the pipeline, the Python-based simulation is converted into C++ code. The comprehensive investigations of neural networks require parallel training. In the talk, an overview of the version 0.2 of the framework ML-PPA is given.