Papers by DEVANSH Agarwal
Searches for Fast Radio Bursts using Machine Learning Devansh Agarwal Fast Radio bursts (FRBs) ar... more Searches for Fast Radio Bursts using Machine Learning Devansh Agarwal Fast Radio bursts (FRBs) are enigmatic astrophysical events with millisecond durations and flux densities in the range 0.1-100 Jy, with the prototype source discovered by Lorimer et al. (2007). Like pulsars, FRBs show the characteristic inverse square sweep in observing frequency due to propagation through an ionized medium. This effect is quantified by the dispersion measure (DM). Unlike pulsars, FRBs have anomalously high DMs, which are consistent with an extragalactic origin. Over 100 FRBs have been published at the time of writing, and 13 have been conclusively identified with host galaxies with spectroscopically determined redshifts in the range 0.003 z 0.66. Detection of FRBs requires data at radio frequencies to be de-dispersed at many trial DM values. Incoming radio telescope data are appropriately combined for each DM to form a time series that is then searched using matched filters to find events above a certain signal-to-noise threshold. In the past, diagnostic plots showing these events are most commonly inspected by humans to determine if they are of astrophysical origin. With ongoing FRB surveys producing millions of candidates, machine learning algorithms for candidate classification are now necessary. In this thesis, we present state-of-the-art deep neural networks to classify FRB candidates and events produced by radio frequency interference (RFI). We present 11 deep learning models named FETCH, each with accuracy and recall above 99.5% as determined using a dataset comprising real RFI and pulsar candidates. These algorithms are telescope and frequency agnostic and can correctly classify all FRBs with signal-to-noise ratios above 10 in datasets collected with the Parkes telescope and the Australian Square Kilometre Array Pathfinder (ASKAP). We present the design, deployment, and initial results from the real-time commensal FRB search pipeline at the Robert C. Byrd Green Bank Telescope (GBT) named greenburst. The pipeline uses FETCH to winnow down the vast number of false-positive single-pulse candidates that mostly result from RFI. In our observations totaling 276 days so far, we have detected individual pulses from 20 known radio pulsars, which provide excellent verification of the system performance. Although no FRBs have been detected to date, we have used our results to update the analysis of Lawrence et al. (2017) to constrain the FRB all-sky rate to be 1140 +200 −180 per day above a peak flux density of 1 Jy. We also constrain the source count index α = 0.84 ± 0.06, substantially flatter than expected from a Euclidean distribution of standard candles (where α = 1.5). We make predictions for detection rates with greenburst as well as other ongoing and planned FRB experiments. Lastly, we present the discovery of FRB 180417 through a targeted search for faint FRBs near the core of the Virgo cluster using ASKAP. Several radio telescopes promptly followed up the FRB for a total of 27 h, but no repeat bursts were detected. An optical follow-up of FRB 180417 using the PROMPT5 telescope revealed no new sources down to an R-band magnitude of 20.1. We argue that FRB 180417 is likely behind the Virgo cluster as the Galactic and intracluster DM contributions are small compared to the DM of the FRB, and there are no galaxies in the line of sight. Adopting an FRB rate of 10 3 FRBs sky −1 day −1 with flux above 1 Jy out to z = 1, our non-detection of FRBs from Virgo constrains (at 68% confidence limit) the faint-end slope of the luminosity function α < 1.6, and the minimum luminosity, L min 6.5 × 10 39 erg s −1 .
With the upcoming commensal surveys for Fast Radio Bursts (FRBs), and their high candidate rate, ... more With the upcoming commensal surveys for Fast Radio Bursts (FRBs), and their high candidate rate, usage of machine learning algorithms for candidate classification is a necessity. Such algorithms will also play a pivotal role in sending real-time triggers for prompt follow-ups with other instruments. In this paper, we have used the technique of Transfer Learning to train the state-of-the-art deep neural networks for classification of FRB and Radio Frequency Interference (RFI) candidates. These are convolutional neural networks which work on radio frequency-time and dispersion measure-time images as the inputs. We trained these networks using simulated FRBs and real RFI candidates from telescopes at the Green Bank Observatory. We present 11 deep learning models, each with an accuracy and recall above 99.5% on our test dataset comprising of real RFI and pulsar candidates. As we demonstrate, these algorithms are telescope and frequency agnostic and are able to detect all FRBs with signa...
Uploads
Papers by DEVANSH Agarwal