Talks

Modern instrumentation is allowing absolutely spectacular new surveys of large portions of the sky for time-domain astrophysical signals such as bursts and pulsations. However, the noise and background environments are also the worst they've ever been (and they aren't going to get better). In this talk I'll go over some tips for how you can make sure that your new super-duper algorithm and code is really finding the astrophysical signals that you think it is or should be. Amongst other things, I'll very briefly cover testing, statistics, injections, and the importance of making your software available to others.

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.
 

RFI mitigation algorithms aim to remove this interference and improve the signal-to-noise ratio of any detectable transient, but with the growing number of techniques, selecting the most appropriate methodology for a given survey can be problematic. Currently, the strategy is decided by the astronomer, which is not scalable for real-time systems planned for next-generation telescopes such as the SKAO telescopes. We have tried to explore the algorithm selection problem by simulating several RFI environments along with pulses injected into a white noise filterbank file, which is then cleaned using each of a number of RFI mitigation algorithms and run through a single pulse search pipeline to analyse the recovery of the injected pulses. We examine the recovery of the known single-pulse parameters with an emphasis on a number of realistic corner cases. In addition to this, look for some solutions to the identified problem as well.

Detecting pulsars in binary systems is crucial as they provide unique laboratories for testing gravitational theories. Traditional methods like Fourier Domain Acceleration Search and Fourier Domain `Jerk' Search, while effective, require significant computation resources and time to correct the effect of Doppler shifts, which limits real-time data processing and the efficiency of discovering binary pulsar systems.

LeoNet addresses this limitation by leveraging the fundamental principles of FDAS and FDJS and incorporating a filter template layer within a convolutional neural network. This layer contains minimal templates, allowing the neural network to optimize the filter parameters.  The filter template layer is divided into two parts, which process the real and imaginary parts of the spectrum, respectively. After convolution, the absolute value of the template layer result is calculated to generate the f_fdot plane for the classification task.

Several subsequent convolution layers and one fully connected layer act as a binary classifier to detect binary pulsar signals. The network continues to reduce the false negative rate while ensuring the false positive rate is 0 on the simulated dataset, reducing the number of binary pulsars that have not been detected. At the same time, it reduces the time required to search for binary pulsar signals. After using TensorRT, the speed of processing signals by the neural network has been improved.

Fast Radio Bursts (FRBs) have emerged as one of the most enigmatic phenomena in modern astrophysics, characterized by their millisecond-duration flashes of extreme luminosity originating from cosmological distances. Despite significant progress in the field, the exact nature and progenitors of FRBs remain elusive. Our group has made substantial advancements in the precise localization of FRBs using Very Long Baseline Interferometry (VLBI), leveraging the unparalleled resolution and sensitivity of the European VLBI Network (EVN). These efforts have enabled us to localize the millisecond-duration bursts to milliarcsecond precision, and directly image the persistent radio sources associated with some FRBs. These findings provided critical insights into their environments and potential progenitors, such as young magnetars, superluminous supernovae, and massive black hole systems. In this talk, I will present the latest results from our VLBI observations within the PRECISE and AstroFlash activities. By combining multi-wavelength data and theoretical modeling, we have been able to differentiate between various formation channels and evolutionary stages of FRB sources. Our findings not only shed light on the physical conditions required to produce these bursts but also pave the way for future research with next-generation facilities. This work represents a significant step forward in understanding the origins and mechanisms of FRBs, contributing to the broader field of astrophysics and cosmology.

In QFT, there exists a well-known set of relations between the scattering amplitudes of gluons in gauge theory and those of gravitons in gravity, known collectively as the double copy. There are many beautiful ways in which this is manifested, famous examples being the KLT double copy, the double copy in the CHY formalism, and the BCJ double copy. In this talk I will show how the double copy relation can be seen arising from amplitude expressions derived using twistor sigma models and the RSVW and Cachazo-Skinner formulae. This talk is based on 2406.04539, with Tim Adamo.