ICTS:32950

A Novel Two-Stage Approach for Robust Single Pulse Sifting

(2025). A Novel Two-Stage Approach for Robust Single Pulse Sifting. SciVideos. https://scivideos.org/index.php/icts-tifr/32950

A Novel Two-Stage Approach for Robust Single Pulse Sifting. SciVideos, Oct. 12, 2025, https://scivideos.org/index.php/icts-tifr/32950 

          @misc{ scivideos_ICTS:32950,
            doi = {},
            url = {https://scivideos.org/index.php/icts-tifr/32950},
            author = {},
            keywords = {},
            language = {en},
            title = {A Novel Two-Stage Approach for Robust Single Pulse Sifting},
            publisher = {},
            year = {2025},
            month = {oct},
            note = {ICTS:32950 see, \url{https://scivideos.org/index.php/icts-tifr/32950}}
          }
Shubham Singh
Talk numberICTS:32950
Source RepositoryICTS-TIFR

Abstract

The single-pulse search in real telescope data can result in a very high candidate rate, going up to a few million candidates per minute. Handling this high candidate rate in a real-time search becomes a big challenge. The step of sifting tries to reduce this candidate rate by clustering detections possibly arising from the same events and rejecting obvious RFI candidates.

I will present a single-pulse sifting pipeline that uses the HDBSCAN clustering to cluster the related candidates, and then a Random Forest classifier on a set of cluster features to identify and reject RFI candidates. The HDBSCAN clustering method is robust against noise and requires only one user-defined parameter, minimizing the possibility of any bias in the final candidates. I will present a set of effective and easy-to-compute cluster features for the Random Forest classifier. I will also discuss ways to minimize the potential loss of astrophysical candidates in this preliminary classification.