Pulsar candidate identification using semi-supervised generative adversarial networks

Balakrishnan, V.; Champion, D. J.; Barr, E. D.; Krämer, M.; Sengar, Rahul; Bailes, M.

doi:10.1093/mnras/stab1308

Cited by 21 publications

(15 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Cameron et al 2018); yet, the most accelerated systems consisting of two neutron stars are typically found at low Galactic latitude. If this is the case, different approaches to searching for pulsars in this data may be required, for example, template matching or other novel search methods (Balakrishnan et al 2021). Moreover, observations with telescopes with greater instantaneous sensitivity could also prove fruitful, although this may not be possible until the SKA-Low is constructed (for sources visible in the southern hemisphere).…”

Section: Non-detection Of Other Polarised Sources As Pulsarsmentioning

confidence: 99%

Searching for pulsars associated with polarised point sources using LOFAR: Initial discoveries from the TULIPP project

Sobey

Bassa

O’Sullivan

et al. 2022

A&A

View full text Add to dashboard Cite

Discovering radio pulsars, particularly millisecond pulsars (MSPs), is important for a range of astrophysical applications, such as testing theories of gravity or probing the magneto-ionic interstellar medium. We aim to discover pulsars that may have been missed in previous pulsar searches by leveraging known pulsar observables (primarily polarisation) in the sensitive, low-frequency radio images from the Low-Frequency Array (LOFAR) Two-metre Sky Survey (LoTSS), and have commenced the Targeted search, using LoTSS images, for polarised pulsars (TULIPP) survey. For this survey, we identified linearly and circularly polarised point sources with flux densities brighter than 2 mJy in LoTSS images at a centre frequency of 144 MHz with a 48 MHz bandwidth. Over 40 known pulsars, half of which are MSPs, were detected as polarised sources in the LoTSS images and excluded from the survey. We have obtained beam-formed LOFAR observations of 30 candidates, which were searched for pulsations using coherent de-dispersion. Here, we present the results of the first year of the TULIPP survey. We discovered two pulsars, PSRs J1049+5822 and J1602+3901, with rotational periods of P = 0.73 s and 3.7 ms, respectively. We also detected a further five known pulsars (two slowly-rotating pulsars and three MSPs) for which accurate sky positions were not available to allow a unique cross-match with LoTSS sources. This targeted survey presents a relatively efficient method by which pulsars, particularly MSPs, may be discovered using the flexible observing modes of sensitive radio telescopes such as the Square Kilometre Array and its pathfinders/precursors, particularly since wide-area all-sky surveys using coherent de-dispersion are currently computationally infeasible.

show abstract

Section: Non-detection Of Other Polarised Sources As Pulsarsmentioning

confidence: 99%

Searching for pulsars associated with polarised point sources using LOFAR: Initial discoveries from the TULIPP project

Sobey

Bassa

O’Sullivan

et al. 2022

A&A

View full text Add to dashboard Cite

show abstract

“…ML has played an important role in developing such new algorithms (Ball & Brunner 2010;Allen et al 2019;Baron 2019;Fluke & Jacobs 2020). For science related to compact objects, ML algorithms have for example been developed to classify new pulsar candidates (Bethapudi & Desai 2018;Lin et al 2020;Balakrishnan et al 2021) as well as transient radio events such as fast radio bursts (Agarwal et al 2020). Other approaches have aimed at forecasting and analyzing gravitational-wave signals in real time (Cabero et al 2020;Gerosa et al 2020;Skliris et al 2020;Wei & Huerta 2020), interpreting gravitational-wave events in light of population synthesis (Wong & Gerosa 2019), or reconstructing the equation of state of a neutron star from observed quantities (Morawski & Bejger 2020).…”

Section: Machine-learning Setupmentioning

confidence: 99%

Analyzing the Galactic Pulsar Distribution with Machine Learning

Ronchi

Graber

García-García

et al. 2021

ApJ

View full text Add to dashboard Cite

We explore the possibility of inferring the properties of the Galactic population of neutron stars through machine learning. In particular, in this paper we focus on their dynamical characteristics and show that an artificial neural network is able to estimate with high accuracy the parameters that control the current positions of a mock population of pulsars. For this purpose, we implement a simplified population-synthesis framework (where selection biases are neglected at this stage) and concentrate on the natal kick-velocity distribution and the distribution of birth distances from the Galactic plane. By varying these and evolving the pulsar trajectories in time, we generate a series of simulations that are used to train and validate a suitably structured convolutional neural network. We demonstrate that our network is able to recover the parameters governing the distribution of kick velocity and Galactic height with a mean relative error of about 10 −2 . We discuss the limitations of our idealized approach and study a toy problem to introduce selection effects in a phenomenological way by incorporating the observed proper motions of 216 isolated pulsars. Our analysis highlights that by increasing the sample of pulsars with accurate proper-motion measurements by a factor of ∼10, one of the future breakthroughs of the Square Kilometre Array, we might succeed in constraining the birth spatial and kick-velocity distribution of the neutron stars in the Milky Way with high precision through machine learning.

show abstract

“…In the practical scenario, there is typically a small amount of labeled data along with a large amount of unlabeled data. Semisupervised learning addresses this imbalance (Balakrishnan et al 2021). Considering that a semisupervised generative adversarial network (SGAN) can learn from unlabeled data of pulsar surveys, Balakrishnan et al (2021) employed an SGAN for pulsar-candidate identification with limited labeled data in the early stages of pulsar surveys on new instruments.…”

Section: Introductionmentioning

confidence: 99%

“…Semisupervised learning addresses this imbalance (Balakrishnan et al 2021). Considering that a semisupervised generative adversarial network (SGAN) can learn from unlabeled data of pulsar surveys, Balakrishnan et al (2021) employed an SGAN for pulsar-candidate identification with limited labeled data in the early stages of pulsar surveys on new instruments. The main advantage of the network is the capacity to leverage readily available unlabeled candidates for achieving excellent results.…”

Section: Introductionmentioning

confidence: 99%

Pulsar-candidate Selection Using a Generative Adversarial Network and ResNeXt

Yin¹,

Li²,

Li³

et al. 2022

ApJS

View full text Add to dashboard Cite

Pulsar research has been a hot topic in the area of astronomy since they were first discovered. Pulsar discovery is fundamental for pulsar research. While pulsars are now visible across the electromagnetic spectrum, pulsar searches with modern radio telescopes are most promising. As the performance of astronomical instruments improves, the number of pulsar candidates detected by modern radio telescopes grows at an exponential rate. The application of artificial intelligence to the field of pulsar-candidate identification can automatically and efficiently address the identification problem with enormous amounts of data. However, there are still significant challenges in enhancing the accuracy of deep-learning-based pulsar-candidate identification. These problems result primarily from the fact that real pulsar data is scarce: the number of candidates that can be successfully identified as real pulsars (positive samples) is much smaller than those candidates that turn out to not be pulsars but instead radio-frequency interference or noise (negative samples). This makes it difficult to train a machine-learning model that can accurately select those candidates that are real pulsars. Therefore a novel pulsar-candidate identification framework is proposed that combines a deep convolutional generative adversarial neural network (DCGAN) and a deep aggregation residual network (ResNeXt). To overcome sample imbalance, the DCGAN is utilized to generate images that approximate real pulsars, while observed and generated candidates are employed together to train the pulsar-candidate identification model ResNeXt. Experiments on the HTRU Medlat data set back up the framework’s performance. The precision, recall, and F1-score of the framework are 100%.

show abstract

Pulsar candidate identification using semi-supervised generative adversarial networks

Cited by 21 publications

References 33 publications

Searching for pulsars associated with polarised point sources using LOFAR: Initial discoveries from the TULIPP project

Searching for pulsars associated with polarised point sources using LOFAR: Initial discoveries from the TULIPP project

Analyzing the Galactic Pulsar Distribution with Machine Learning

Pulsar-candidate Selection Using a Generative Adversarial Network and ResNeXt

Contact Info

Product

Resources

About