Oral Presentation

Machine learning classification of baseband data of CHIME FRBs

Author(s): Mohanraj Madheshwaran (Department of Physics, National Chung Hsing University, Taichung, Taiwan) Tetsuya Hashimoto (Department of Physics, National Chung Hsing University, Taichung, Taiwan) Tomotsugu Goto (Institute of Astronomy, National Tsing Hua University, Hsinchu, Taiwan) Simon C.-C. Ho (Research School of Astronomy and Astrophysics, The Australian National University, Canberra, ACT 2611, Australia) Vignesh Vavillakula venkataramana Rao (Department of Physics, National Chung Hsing University, Taichung, Taiwan) Murthadza Aznam (Department of Physics, Faculty of Science, Universiti Malaya, Kuala Lumpur 50603, Malaysia.) Sridhar Gajendran (Institute of Astronomy, National Tsing Hua University, Hsinchu, Taiwan)

Presenter: Mohanraj Madheshwaran (National Chung Hsing University)

Fast Radio Bursts (FRBs) are brief, millisecond-duration radio pulses of unknown origin. While some FRBs are known to repeat, others appear as one-off events. Repeaters often exhibit broader pulses and narrower bandwidths compared to non-repeaters, suggesting potentially distinct origins. Accurate classification of FRBs is essential but remains challenging due to the limited monitoring capabilities and sensitivity of current radio telescopes. Consequently, some faint repeating FRBs may be misclassified as non-repeaters—these are referred to as repeater candidates.
To address this, machine learning has emerged as a promising tool. Previous studies applied uniform manifold approximation and projection (UMAP) to the CHIME/FRB Catalog 1 and identified 188 repeater candidates among 474 non-repeaters. In this work, we analyze the CHIME/FRB baseband catalog, which offers significant improvements in positional accuracy (~10 arcsec) and time resolution for 140 FRBs. These enhancements provide more reliable measurements of fluence, duration, and position, allowing us to refine classification methods and re-evaluate prior findings.
Our objectives are: (i) to assess consistency with previous candidate identifications, (ii) to identify new repeater candidates, and (iii) to analyze the relationship between current and past results. We identified 15 repeater candidates among 122 non-repeating FRBs. Of these, 14 match candidates reported previously, while one is newly identified. Notably, one of our candidates has since been confirmed as a repeater by the CHIME/FRB Collaboration, validating our approach.