Machine-learning Classification of FRB 20121102A using Arecibo Data
Tetsuya Hashimoto (National Chung Hsing University)
Description:
Fast radio bursts are mysterious radio pulses with millisecond durations. The classification of FRBs is fundamental because it could provide us with a hint to investigate their emission mechanisms. However, previous classification schemes relied on visual inspections or used only a few parameters, suffering from possible human bias. To overcome this problem, we use an unsupervised machine-learning approach to classify an actively repeating FRB, FRB 20121102A.
More than 800 repeating FRBs are detected from this source with the Arecibo radio telescope. Using the ~800 FRBs, we use Uniform Manifold Approximation and Projection (UMAP) to project the multi-dimension data onto a plane. UMAP allows us to handle eight observed parameters of the ~800 FRBs, simultaneously. This projection will allow us to optimize the clustering algorism to find possible groups in the samples. Because the different groups might correspond to different emission mechanisms, we will investigate the characteristic physical properties of each group. This project will shed light on the possible FRB mechanisms, being free of human bias.