High Time Resolution Universe Research Articles

The High Time Resolution Universe Pulsar Survey – XIX. A coherent GPU-accelerated reprocessing and the discovery of 71 pulsars in the Southern Galactic plane

ABSTRACT We conducted a GPU-accelerated reprocessing of $\sim 87~{{\ \rm per\ cent}}$ of the archival data from the High Time Resolution Universe South Low Latitude (HTRU-S LowLat) pulsar survey by implementing a pulsar search pipeline that was previously used to reprocess the Parkes Multibeam Pulsar Survey (PMPS). We coherently searched the full 72-min observations of the survey with an acceleration search range up to $|50|\, \rm m\, s^{-2}$, which is most sensitive to binary pulsars experiencing nearly constant acceleration during 72 min of their orbital period. Here we report the discovery of 71 pulsars, including six millisecond pulsars, of which five are in binary systems, and seven pulsars with very high dispersion measures (DM $\gt 800 \, \rm pc \, cm^{-3}$). These pulsar discoveries largely arose by folding candidates to a much lower spectral signal-to-noise ratio than in previous surveys and by exploiting the coherence of folding over the incoherent summing of the Fourier components to discover new pulsars as well as candidate classification techniques. We show that these pulsars could be fainter and on average more distant as compared with both the previously reported 100 HTRU-S LowLat pulsars and the background pulsar population in the survey region. We have assessed the effectiveness of our search method and the overall pulsar yield of the survey. We show that through this reprocessing we have achieved the expected survey goals, including the predicted number of pulsars in the survey region, and discuss the major causes why these pulsars were missed in previous processing of the survey.

Eighteen new fast radio bursts in the High Time Resolution Universe survey

Context. Current observational evidence reveals that fast radio bursts (FRBs) exhibit bandwidths ranging from a few dozen MHz to several GHz. Traditional FRB searches primarily employ matched filter methods on time series collapsed across the entire observational bandwidth. However, with modern ultrawideband receivers featuring gigahertz-scale observational bandwidths, this approach may overlook a significant number of events. Aims. We investigate the efficacy of sub-banded searches for FRBs, whereby we look for bursts within limited portions of the bandwidth. The aim of these searches is to enhance the significance of FRB detections by mitigating the impact of noise outside the targeted frequency range, thereby improving signal-to-noise ratios. Methods. We conducted a series of Monte Carlo simulations for the 400-MHz bandwidth Parkes 21-cm multi-beam (PMB) receiver system and the Parkes Ultra-Wideband Low (UWL) receiver, simulating bursts down to frequency widths of about 100 MHz. Additionally, we performed a complete reprocessing of the high-latitude segment of the High Time Resolution Universe South survey (HTRU-S) of the Parkes-Murriyang telescope using sub-banded search techniques. Results. Simulations reveal that a sub-banded search can enhance the burst search efficiency by 67−42+133% for the PMB system and 1433−126+143% for the UWL receiver. Furthermore, the reprocessing of HTRU led to the confident detection of 18 new bursts, nearly tripling the count of FRBs found in this survey. Conclusions. These results underscore the importance of employing sub-banded search methodologies to effectively address the often modest spectral occupancy of these signals.

Pulsar candidate identification using advanced transformer-based models

Rapid and accurate identification of pulsars is a significant topic for large radio telescope surveys. With the enhancement of astronomical instruments, modern radio telescopes are witnessing an exponential increase in pulsar candidate detections. The application of artificial intelligence for the identification of pulsar candidates is an automated and highly effective solution to tackle the challenge of processing and recognizing vast volumes of data. In this work, using the data released by two surveys, the Commensal Radio Astronomy FasT Survey (CRAFTS) and High-Time Resolution Universe (HTRU), we propose a new framework to identify pulsar candidates. Firstly, due to the small number of real pulsars, we compare the performance of different data augmentation methods and find that the pulsar samples generated by the Deep Convolutional Generative Adversarial Network (DCGAN) based on deep learning techniques are closer to real pulsars. Secondly, we use two transformer-based classification models, Vision Transformer (ViT) and Convolutional Vision Transformer (CvT), to classify pulsar candidates, and find that the evaluation indexes of pulsar candidate classification based on two transformers can reach 100%. Finally, we use the t-distributed Stochastic Neighbor Embedding (t-SNE) algorithm to visualize the results of our identification framework. The results showed that pulsar and non-pulsar samples are separated from each other in multidimensional space. Therefore, it is a new attempt to apply transformer technology to pulsar candidate classification, and it could be of great significance to subsequent theoretical research.

Applying Hybrid Clustering in Pulsar Candidate Sifting with Multi-modality for FAST Survey

Pulsar search is always the basis of pulsar navigation, gravitational wave detection and other research topics. Currently, the volume of pulsar candidates collected by the Five-hundred-meter Aperture Spherical radio Telescope (FAST) shows an explosive growth rate that has brought challenges for its pulsar candidate filtering system. Particularly, the multi-view heterogeneous data and class imbalance between true pulsars and non-pulsar candidates have negative effects on traditional single-modal supervised classification methods. In this study, a multi-modal and semi-supervised learning based on a pulsar candidate sifting algorithm is presented, which adopts a hybrid ensemble clustering scheme of density-based and partition-based methods combined with a feature-level fusion strategy for input data and a data partition strategy for parallelization. Experiments on both High Time Resolution Universe Survey II (HTRU2) and actual FAST observation data demonstrate that the proposed algorithm could excellently identify pulsars: On HTRU2, the precision and recall rates of its parallel mode reach 0.981 and 0.988 respectively. On FAST data, those of its parallel mode reach 0.891 and 0.961, meanwhile, the running time also significantly decreases with the increment of parallel nodes within limits. Thus, we can conclude that our algorithm could be a feasible idea for large scale pulsar candidate sifting for FAST drift scan observation.

Dealing with the Data Imbalance Problem in Pulsar Candidate Sifting Based on Feature Selection

Pulsar detection has become an active research topic in radio astronomy recently. One of the essential procedures for pulsar detection is pulsar candidate sifting (PCS), a procedure for identifying potential pulsar signals in a survey. However, pulsar candidates are always class-imbalanced, as most candidates are non-pulsars such as RFI and only a tiny part of them are from real pulsars. Class imbalance can greatly affect the performance of machine learning (ML) models, resulting in a heavy cost as some real pulsars are misjudged. To deal with the problem, techniques of choosing relevant features to discriminate pulsars from non-pulsars are focused on, which is known as feature selection. Feature selection is a process of selecting a subset of the most relevant features from a feature pool. The distinguishing features between pulsars and non-pulsars can significantly improve the performance of the classifier even if the data are highly imbalanced. In this work, an algorithm for feature selection called the K-fold Relief-Greedy (KFRG) algorithm is designed. KFRG is a two-stage algorithm. In the first stage, it filters out some irrelevant features according to their K-fold Relief scores, while in the second stage, it removes the redundant features and selects the most relevant features by a forward greedy search strategy. Experiments on the data set of the High Time Resolution Universe survey verified that ML models based on KFRG are capable of PCS, correctly separating pulsars from non-pulsars even if the candidates are highly class-imbalanced.

The High Time Resolution Universe Pulsar survey – XVIII. The reprocessing of the HTRU-S Low Lat survey around the Galactic Centre using a Fast Folding Algorithm pipeline for accelerated pulsars

ABSTRACT The HTRU-S Low Latitude survey data within 1° of the Galactic Centre (GC) were searched for pulsars using the Fast Folding Algorithm (FFA). Unlike traditional Fast Fourier Transform (FFT) pipelines, the FFA optimally folds the data for all possible periods over a given range, which is particularly advantageous for pulsars with low-duty cycles. For the first time, a search over acceleration was included in the FFA to improve its sensitivity to binary pulsars. The steps in dispersion measure (DM) and acceleration were optimized, resulting in a reduction of the number of trials by 86 per cent. This was achieved over a search period range from 0.6 to 432-s, i.e. 10 per cent of the observation time (4320s), with a maximum DM of 4000 pc cm−3 and an acceleration range of ±128 m s−2. The search resulted in the re-detections of four known pulsars, including a pulsar that was missed in the previous FFT processing of this survey. This result indicates that the FFA pipeline is more sensitive than the FFT pipeline used in the previous processing of the survey within our parameter range. Additionally, we discovered a 1.89-s pulsar, PSR J1746-2829, with a large DM, located 0.5 from the GC. Follow-up observations revealed that this pulsar has a relatively flat spectrum (α = −0.9 ± 0.1) and has a period derivative of ∼1.3 × 10−12 s s−1, implying a surface magnetic field of ∼5.2 × 1013 G and a characteristic age of ∼23 000 yr. While the period, spectral index, and surface magnetic field strength are similar to many radio magnetars, other characteristics such as high linear polarization are absent.

Deep learning-based pulsar candidate identification model using a variational autoencoder

As radio telescope technology continues to advance, the discovery of more pulsar candidates is anticipated. The swift and accurate identification of pulsars has emerged as a compelling research topic in recent years, garnering growing attention. Although deep learning has been expansively applied for the rapid identification of pulsar candidates, the significant imbalance between the larger number of non-pulsar signal samples and the fewer pulsar signal samples seriously impacts the performance of deep learning prediction models. To address this, our research proposes a method that combines a Vector Quantized Variational Autoencoder (VQVAE) with a Gated PixelCNN to equalize the number of pulsar and non-pulsar samples during sample generation. Subsequently, we establish a Residual Neural Network (ResNet) model that employs two feature diagnostic plots as input to classify and identify pulsar candidates. Validation on the High Time Resolution Universe (HTRU) dataset yielded a recall and accuracy of 99.60%, and an F1-score of 99.59%. Further evaluation on the Five-hundred-meter Aperture Spherical radio Telescope (FAST) dataset demonstrated a precision of 98.5%, a recall of 97.8%, and an F1-score of 98.1%, further confirming the effectiveness of our proposed model.

MFPIM: A Deep Learning Model Based on Multimodal Fusion Technology for Pulsar Identification

With the development of radio telescope technology, the quantity and types of acquired pulsar candidate data have increased dramatically. However, it is difficult to accurately identify pulsar candidates. Therefore, we propose to use multimodal fusion technology, called the multimodal fusion-based pulsar identification model (MFPIM), to build a deep learning model to improve the efficiency and accuracy of pulsar candidate identification. MFPIM treats each diagnostic plot of pulsar candidates as a modality and uses multiple convolutional neural networks to extract effective features from the diagnostic plots. After fusing the features, the commonality of different modalities in high-dimensional space is obtained to ensure that the model can take full advantage of the complementarity between diagnostic plots and thus identify pulsar candidates, achieved better classification performance than other current supervised learning algorithms. In addition, a channel attention mechanism is used in the model to enable it to learn the importance of different channel features so that the model focuses more on the channel information in the input data that is more meaningful for classification, reducing the model size while extracting pulsar diagnostic map features more accurately. We conducted experiments on the Five-hundred-meter Aperture Spherical radio Telescope (FAST) data set, and the results show that MFPIM can effectively identify the pulsars in the FAST data set with an identification accuracy of over 98%. To further verify the robustness of the model, we applied the MFPIM to the High Time Resolution Universe data set using transfer learning, with the test accuracy and F1 score reaching over 99%.

The Southern-sky MWA Rapid Two-metre (SMART) pulsar survey—II. Survey status, pulsar census, and first pulsar discoveries

Abstract In Paper I, we presented an overview of the Southern-sky MWA Rapid Two-metre (SMART) survey, including the survey design and search pipeline. While the combination of MWA’s large field-of-view and the voltage capture system brings a survey speed of ${\sim} 450\, {\textrm{deg}}^{2}\,\textrm{h}^{-1}$ , the progression of the survey relies on the availability of compact configuration of the Phase II array. Over the past few years, by taking advantage of multiple windows of opportunity when the compact configuration was available, we have advanced the survey to 75% of the planned sky coverage. To date, about 10% of the data collected thus far have been processed for a first-pass search, where 10 min of observation is processed for dispersion measures out to 250 ${\textrm{pc cm}}^{-3}$ , to realise a shallow survey that is largely sensitive to long-period pulsars. The ongoing analysis has led to two new pulsar discoveries, as well as an independent discovery and a rediscovery of a previously incorrectly characterised pulsar, all from ${\sim} 3\% $ of the data for which candidate scrutiny is completed. In this sequel to Paper I, we describe the strategies for further detailed follow-up including improved sky localisation and convergence to timing solution, and illustrate them using example pulsar discoveries. The processing has also led to re-detection of 120 pulsars in the SMART observing band, bringing the total number of pulsars detected to date with the MWA to 180, and these are used to assess the search sensitivity of current processing pipelines. The planned second-pass (deep survey) processing is expected to yield a three-fold increase in sensitivity for long-period pulsars, and a substantial improvement to millisecond pulsars by adopting optimal de-dispersion plans. The SMART survey will complement the highly successful Parkes High Time Resolution Universe survey at 1.2–1.5 GHz, and inform future large survey efforts such as those planned with the low-frequency Square Kilometre Array (SKA-Low).

A Preliminary Study of Large Scale Pulsar Candidate Sifting Based on Parallel Hybrid Clustering

Pulsar candidate sifting is an essential part of pulsar analysis pipelines for discovering new pulsars. To solve the problem of data mining of a large number of pulsar data using a Five-hundred-meter Aperture Spherical radio Telescope (FAST), a parallel pulsar candidate sifting algorithm based on semi-supervised clustering is proposed, which adopts a hybrid clustering scheme based on density hierarchy and the partition method, combined with a Spark-based parallel model and a sliding window-based partition strategy. Experiments on the two datasets, HTRU (The High Time-Resolution Universe Survey) 2 and AOD-FAST (Actual Observation Data from FAST), show that the algorithm can excellently identify the pulsars with high performance: On HTRU2, the Precision and Recall rates are 0.946 and 0.905, and those on AOD-FAST are 0.787 and 0.994, respectively; the running time on both datasets is also significantly reduced compared with its serial execution mode. It can be concluded that the proposed algorithm provides a feasible idea for astronomical data mining of FAST observation.

The High Time Resolution Universe Pulsar Survey – XVII. PSR J1325−6253, a low eccentricity double neutron star system from an ultra-stripped supernova

ABSTRACT The observable population of double neutron star (DNS) systems in the Milky Way allow us to understand the nature of supernovae and binary stellar evolution. Until now, all DNS systems in wide orbits (Porb > 1 d) have been found to have orbital eccentricities, e > 0.1. In this paper, we report the discovery of pulsar PSR J1325−6253: a DNS system in a 1.81-d orbit with a surprisingly low eccentricity of just e = 0.064. Through, 1.4 yr of dedicated timing with the Parkes radio telescope we have been able to measure its rate of advance of periastron, $\dot{\omega }=0{_{.}^{\circ}}138\pm 0{_{.}^{\circ}}002\, \mathrm{yr}^{-1}$. If this induced $\dot{\omega }$ is solely due to general relativity then the total mass of the system is, Msys = 2.57 ± 0.06 M⊙. Assuming an edge-on orbit the minimum companion mass is constrained to be Mc, min > 0.98 M⊙ which implies the pulsar mass is Mp, max < 1.59 M⊙. Its location in the P–$\dot{P}$ diagram suggests that, like other DNS systems, PSR J1325−6253 is a recycled pulsar and if its mass is similar to the known examples (>1.3 M⊙), then the companion neutron star is probably less than ∼1.25 M⊙ and the system is inclined at about 50°–60°. The low eccentricity along with the wide orbit of the system strongly favours a formation scenario involving an ultra-stripped supernova explosion.

AdaBoost-MICNN: a new network framework for pulsar candidate selection

ABSTRACT Pulsar observation and research are of great significance. With the gradual increase in the performance and quantity of observing equipment, the received pulsar observation data also increase geometrically. Machine learning can mine large-scale pulsar data sets to realize automatic pulsar candidate identification. However, the pulsar candidate and non-pulsar candidate data sets are highly imbalanced. The traditional classifier trained on the data set with imbalanced class distribution usually shows poor generalization performance, which will cause the classifier to be biased towards negative instances. To address the problem of poor identification algorithm performance caused by sample imbalance, we describe a new framework, namely AdaBoost-multi-input-CNN (AdaBoost-MICNN). In AdaBoost, multiple baseline models are trained sequentially, and the weight of each training sample changes as the training progresses. Especially the minority samples, which are more easily ignored, will get more attention in subsequent baseline models. This will solve the problem of imbalanced classification. Meanwhile, there are usually four pulsar diagnostic plots. To automatically extract and integrate these four diagnostic plots features, we use multi-input-CNN (MICNN) as the baseline model of AdaBoost. The attention mechanism is introduced to enhance the performance of MICNN’s extraction of features, and design an objective function that adds the maximum mean discrepancy to enhance the anti-interference ability of the model. Finally, the proposed method is tested in the High Time Resolution Universe Medlat Data, and the experimental results verify the effectiveness and efficiency of the method.

Classification of pulsar signals using ensemble gradient boosting algorithms based on asymmetric under-sampling method

The detection of pulsar signals is a highly intensive task. Numerous artificial intelligence (AI) and machine learning techniques (ML) have been proposed to classify pulsar and non-pulsar signals. While existing machine learning techniques improve classification efficiency, these methods are limited when it comes to dealing with large volumes of astronomical data, the extreme problem of class imbalance , and the polarization of high recall and precision. In this paper, to accurately classify pulsar and non-pulsar signals, extreme Gradient Boosting (XgBoost), and Light Gradient Boosting Machine (LightGBM) algorithms based on an asymmetric undersampling method are proposed. Firstly, the proposed model uses an asymmetric undersampling technique that divides the benchmark datasets into 75 subsets for the LOFAR Tied-Array All-Sky Survey (LOTAAS-1) and 9 subsets for the High Time Resolution Universe Pulsar Survey (HTRU2), eluding the class imbalance problem. Finally, XgBoost and LightGBM algorithms have been adopted for each new data subset, with a majority voting classifier being used to integrate the output of proposed models against each subset. Results of the proposed method were compared to state-of-the-art baseline models, which showed that our models significantly performed better than existing methods and gained almost 2% and 2.5% better F-score and precision results, respectively.

Adaboost-DSNN: an adaptive boosting algorithm based on deep self normalized neural network for pulsar identification

ABSTRACT A modern pulsar survey generates a large number of pulsar candidates. Filtering these pulsar candidates in a large astronomical data set is an important step towards discovering new pulsars. In this paper, a novel adaptive boosting algorithm based on deep self normalized neural network (Adaboost-DSNN) is proposed to accurately classify pulsar and non-pulsar signals. To train the proposed method on a highly imbalanced data set, the Synthetic Minority Oversampling TEchnique (SMOTE) was initially employed for balancing the data set. Then, a deep ensemble network combined with a deep self-normalized neural network and adaptive boosting was developed to train and learn the processed pulsar data. The design of the proposed Adaboost-DSNN method significantly reduced the computational time when dealing with large astronomical data sets, while also improving the classification performance. The scaled exponential liner units activation function was used to normalize the data. Considering their neighbour information and the special dropout technique (α-dropout), Adaboost-DSNN displayed good pulsar classification performance, while preserving the data properties across subsequent layers. The proposed Adaboost-DSNN method was tested on the High Time Resolution Universe Survey data sets (HTRU-1 and HTRU-2). According to experimental results, Adaboost-DSNN outperform other state-of-the-art methods with respect to training time and F1-score. The training time of the Adaboost-DSNN model is 10x times faster compared to other models of this kind.

Coherent search for binary pulsars across all Five Keplerian parameters in radio observations using the template-bank algorithm

ABSTRACT Relativistic binary pulsars orbiting white dwarfs and neutron stars have already provided excellent tests of gravity. However, despite observational efforts, a pulsar orbiting a black hole has remained elusive. One possible explanation is the extreme Doppler smearing caused by the pulsar’s orbital motion which changes its apparent spin frequency during an observation. The classical solution to this problem has been to assume a constant acceleration/jerk for the entire observation. However, this assumption breaks down when the observation samples a large fraction of the orbit. This limits the length of search observations, and hence their sensitivity. This provides a strong motivation to develop techniques that can find compact binaries in longer observations. Here, we present a GPU-based radio pulsar search pipeline that can perform a coherent search for binary pulsars by directly searching over three or five Keplerian parameters using the template-bank algorithm. We compare the sensitivity obtained from our pipeline with acceleration and jerk search pipelines for simulated pulsar-stellar-mass black hole binaries and observations of PSR J0737−3039A. We also discuss the computational feasibility of our pipeline for untargeted pulsar surveys and targeted searches. Our benchmarks indicate that circular orbit searches for P-BH binaries with spin-period $P{_\mathrm{{spin}}} \ge {20}\, \mathrm{m\, s}$ covering the 3–10 Tobs regime are feasible for the High Time Resolution Universe pulsar survey. Additionally, an elliptical orbit search in Globular clusters for $P{_\mathrm{{spin}}} \ge {20}\, \mathrm{m\, s}$ pulsars orbiting intermediate-mass black holes in the 5–10 Tobs regime is feasible for observations shorter than 2 h with an eccentricity limit of 0.1.

Early-fusion based pulsar identification with smart under-sampling

The discovery of pulsars is of great significance in the field of physics and astronomy. As the astronomical equipment produces a large number of pulsar data, an algorithm for automatically identifying pulsars becomes urgent. We propose a deep learning framework for pulsar recognition. In response to the extreme imbalance between positive and negative examples and the hard negative sample issue presented in the High Time Resolution Universe Medlat Training Data, there are two coping strategies in our framework: the smart under-sampling and the improved loss function. We also apply the early-fusion strategy to integrate features obtained from different attributes before classification to improve the performance. To our best knowledge, this is the first study that integrates these strategies and techniques in pulsar recognition. The experiment results show that our framework outperforms previous works with respect to either the training time or F1 score. We can not only speed up the training time by 10 × compared with the state-of-the-art work, but also get a competitive result in terms of F1 score.

Research on Pulsar Candidate Identification Method Based on Deep Residual Neural Network

As the next-generation radio astronomical telescopes continuously improve and develop, the pulsar survey will produce millions of pulsar candidates, which pose considerable challenges for pulsar identification and classification. The rapidly evolving artificial intelligence (AI) techniques are being used for pulsar identification and discovery of new pulsars. Using the pulsar data set observed with the Parkes telescope, namely the High Time Resolution Universe Survey (HTRUS), a 14-layer deep residual network has been designed (called the Residual Network, ResNet) for pulsar candidate classifications. In the HTRUS sample data, the number of non-pulsar candidates (i.e., negative samples) is much larger than that of pulsar candidates (i.e., positive samples). The imbalance between the positive and negative samples is prone to result in model misjudgement. By using the over-sampling technique to enhance the data of positive samples in the training set and adjusting the ratio of positive and negative samples, we have solved this imbalance problem. In the training process, the hyperparameters are adjusted by means of 5-fold cross validation to build the model. The test results indicate that the model can achieve a high precision (98%) and recall (100%), the F1-score can reach 99%, and that the implementation of each sample test needs only 7 ms, it has provided a feasible approach for the future big-data analysis of pulsars.

Pulsar candidate identification using semi-supervised generative adversarial networks

ABSTRACT Machine learning methods are increasingly helping astronomers identify new radio pulsars. However, they require a large amount of labelled data, which is time consuming to produce and biased. Here, we describe a Semi-supervised generative adversarial network, which achieves better classification performance than the standard supervised algorithms using majority unlabelled data sets. We achieved an accuracy and mean F-Score of 94.9 per cent trained on only 100 labelled candidates and 5000 unlabelled candidates compared to our standard supervised baseline which scored at 81.1 per cent and 82.7 per cent, respectively. Our final model trained on a much larger labelled data set achieved an accuracy and mean F-score value of 99.2 per cent and a recall rate of 99.7 per cent. This technique allows for high-quality classification during the early stages of pulsar surveys on new instruments when limited labelled data are available. We open-source our work along with a new pulsar-candidate data set produced from the High Time Resolution Universe – South Low Latitude Survey. This data set has the largest number of pulsar detections of any public data set and we hope it will be a valuable tool for benchmarking future machine learning models.

Learning from imbalanced pulsar data by combine DCGAN and PILAE algorithm

A pulsar is a rapidly rotating neutron star and transmits periodic oscillations of power to the earth. We introduce a novel method for pulsar candidate classification. The method contains two major steps: (1) make strong representations for pulsar candidate in the image domain by extracting deep features with the deep convolutional generative adversarial Networks (DCGAN) and (2) develop a classifier defined by multilayer perceptron (MLP) neural networks trained with pseudoinverse learning autoencoder (PILAE) algorithm. We utilized the synthetic minority over-sampling technique (SMOTE) to handle the imbalance in the dataset. We report a variety of measure scores from the output of the PILAE method on datasets utilized in the experiments. The PILAE training process does not have to determine the learning control parameters or indicate the number of hidden layers. Therefore, the PILAE classifier can fulfil superior execution in terms of training effectiveness and accuracy. Empirical results from the high time resolution universe (HTRU) mid-latitude dataset, MNIST dataset and CIFAR-10 have demonstrated that the presented framework achieves excellent results with other models and reasonably low complexly.

Pulsar Candidate Sifting Using Multi-input Convolution Neural Networks

Abstract Pulsar candidate sifting is an essential process for discovering new pulsars. It aims to search for the most promising pulsar candidates from an all-sky survey, such as the High Time Resolution Universe (HTRU), Green Bank Northern Celestial Cap (GBNCC), Five-hundred-meter Aperture Spherical Radio Telescope, etc. Recently, machine learning (ML) has become a hot topic in investigations of pulsar candidate sifting. However, one typical challenge in ML for pulsar candidate sifting comes from the learning difficulty arising from the high class imbalance between the observed numbers of pulsars and non-pulsars. Therefore, this work proposes a novel framework for candidate sifting, named Multi-input Convolutional Neural Networks (MICNN). MICNN is an architecture of deep learning with four diagnostic plots of a pulsar candidate as its inputs. To train our MICNN on a highly class-imbalanced data set, a novel image augmentation technique is proposed, as well as a three-stage training strategy. Experiments on observations from HTRU and GBNCC show the effectiveness and robustness of these proposed techniques. In the experiments on HTRU, our MICNN model achieves a recall rate of 0.962 and a precision rate of 0.967 even in a highly class-imbalanced test data set.