Speaker
Description
Our work proposes an optimized pulsar search pipeline that utilizes deep learning to sift ``snapshot'' candidates generated by folding de-dispersed time series data. This approach significantly accelerates the search process. We also developed a script to generate simulated pulsar signals, optimizing the training set and improving model performance. The benchmark uses the globular cluster NGC 5904 data and simulated pulsar data. This approach shows that reducing candidate folding time by a factor of $\sim$10 still maintains full detection capability for all identifiable pulsars. We tested the artificial intelligence (AI) model's pulsar classification on real data collected from the Five-hundred-meter Aperture Spherical radio Telescope (FAST), Green Bank Telescope (GBT), Murchison Widefield Array (MWA), Arecibo, and Parkes (Murriyang), demonstrating that the method can be generalized to different telescopes. Due to the wide-field of view of radio telescope arrays, such as the SKA-Low, even with the long observation times for the pulsar survey, it is possible to complete the observational pulsar survey plan in a short amount of time compared to single-dish telescopes. As the observation time for a single point is extended, the resulting raw data size increases correspondingly. This directly increases the time required for folding the growing raw data. Note that our developed method effectively accelerates pulsar searches if folding dominates computational time.
