arXiv subjects

Anna M. M. Scaife

Publications and source records attributed to Anna M. M. Scaife.

At least 37 records · Page 2Linked to original sources

arXiv2022

Learning useful representations for radio astronomy "in the wild" with contrastive learning

Unknown class distributions in unlabelled astrophysical training data have previously been shown to detrimentally affect model performance due to dataset shift between training and validation sets. For radio galaxy classification, we demonstrate in this work that removing low angular extent sources from the unlabelled data before training produces qualitatively different training dynamics for a contrastive model. By applying the model on an unlabelled data-set with unknown class balance and sub-population distribution to generate a representation space of radio galaxies, we show that with an appropriate cut threshold we can find a representation with FRI/FRII class separation approaching that of a supervised baseline explicitly trained to separate radio galaxies into these two classes. Furthermore we show that an excessively conservative cut threshold blocks any increase in validation accuracy. We then use the learned representation for the downstream task of performing a similarity search on rare hybrid sources, finding that the contrastive model can reliably return semantically similar samples, with the added bonus of finding duplicates which remain after pre-processing.

astro-ph.GA↗

arXiv2022

A GPU-based Imager with Polarised Primary-beam Correction

The next generation of radio telescopes will strive for unprecedented dynamic range across wide fields of view, and direction-dependent gains such as the gain from the primary-beam pattern, or leakage of one Stokes product into another, must be removed from the cleaned images if dynamic range is to reach its full potential. Unfortunately, such processing is extremely computationally intensive, and is made even more challenging by the very large volumes of data that these instruments will generate. Here we describe a new GPU-based imager, aimed primarily at use with the ASKAP telescope, that is capable of generating cleaned, full-polarisation images that include wide-field, primary-beam, and polarisation leakage corrections.

astro-ph.IM↗

arXiv2022

Towards Galaxy Foundation Models with Hybrid Contrastive Learning

New astronomical tasks are often related to earlier tasks for which labels have already been collected. We adapt the contrastive framework BYOL to leverage those labels as a pretraining task while also enforcing augmentation invariance. For large-scale pretraining, we introduce GZ-Evo v0.1, a set of 96.5M volunteer responses for 552k galaxy images plus a further 1.34M comparable unlabelled galaxies. Most of the 206 GZ-Evo answers are unknown for any given galaxy, and so our pretraining task uses a Dirichlet loss that naturally handles unknown answers. GZ-Evo pretraining, with or without hybrid learning, improves on direct training even with plentiful downstream labels (+4% accuracy with 44k labels). Our hybrid pretraining/contrastive method further improves downstream accuracy vs. pretraining or contrastive learning, especially in the low-label transfer regime (+6% accuracy with 750 labels).

cs.CV↗

arXiv2022

Practical Galaxy Morphology Tools from Deep Supervised Representation Learning

Astronomers have typically set out to solve supervised machine learning problems by creating their own representations from scratch. We show that deep learning models trained to answer every Galaxy Zoo DECaLS question learn meaningful semantic representations of galaxies that are useful for new tasks on which the models were never trained. We exploit these representations to outperform several recent approaches at practical tasks crucial for investigating large galaxy samples. The first task is identifying galaxies of similar morphology to a query galaxy. Given a single galaxy assigned a free text tag by humans (e.g. "#diffuse"), we can find galaxies matching that tag for most tags. The second task is identifying the most interesting anomalies to a particular researcher. Our approach is 100% accurate at identifying the most interesting 100 anomalies (as judged by Galaxy Zoo 2 volunteers). The third task is adapting a model to solve a new task using only a small number of newly-labelled galaxies. Models fine-tuned from our representation are better able to identify ring galaxies than models fine-tuned from terrestrial images (ImageNet) or trained from scratch. We solve each task with very few new labels; either one (for the similarity search) or several hundred (for anomaly detection or fine-tuning). This challenges the longstanding view that deep supervised methods require new large labelled datasets for practical use in astronomy. To help the community benefit from our pretrained models, we release our fine-tuning code Zoobot. Zoobot is accessible to researchers with no prior experience in deep learning.

astro-ph.GA↗

arXiv2022

Radio Galaxy Zoo: Using semi-supervised learning to leverage large unlabelled data-sets for radio galaxy classification under data-set shift

In this work we examine the classification accuracy and robustness of a state-of-the-art semi-supervised learning (SSL) algorithm applied to the morphological classification of radio galaxies. We test if SSL with fewer labels can achieve test accuracies comparable to the supervised state-of-the-art and whether this holds when incorporating previously unseen data. We find that for the radio galaxy classification problem considered, SSL provides additional regularisation and outperforms the baseline test accuracy. However, in contrast to model performance metrics reported on computer science benchmarking data-sets, we find that improvement is limited to a narrow range of label volumes, with performance falling off rapidly at low label volumes. Additionally, we show that SSL does not improve model calibration, regardless of whether classification is improved. Moreover, we find that when different underlying catalogues drawn from the same radio survey are used to provide the labelled and unlabelled data-sets required for SSL, a significant drop in classification performance is observered, highlighting the difficulty of applying SSL techniques under dataset shift. We show that a class-imbalanced unlabelled data pool negatively affects performance through prior probability shift, which we suggest may explain this performance drop, and that using the Frechet Distance between labelled and unlabelled data-sets as a measure of data-set shift can provide a prediction of model performance, but that for typical radio galaxy data-sets with labelled sample volumes of O(1000), the sample variance associated with this technique is high and the technique is in general not sufficiently robust to replace a train-test cycle.

astro-ph.GA↗

arXiv2022

Can semi-supervised learning reduce the amount of manual labelling required for effective radio galaxy morphology classification?

In this work, we examine the robustness of state-of-the-art semi-supervised learning (SSL) algorithms when applied to morphological classification in modern radio astronomy. We test whether SSL can achieve performance comparable to the current supervised state of the art when using many fewer labelled data points and if these results generalise to using truly unlabelled data. We find that although SSL provides additional regularisation, its performance degrades rapidly when using very few labels, and that using truly unlabelled data leads to a significant drop in performance.

astro-ph.GA↗

arXiv2022

Quantifying Uncertainty in Deep Learning Approaches to Radio Galaxy Classification

In this work we use variational inference to quantify the degree of uncertainty in deep learning model predictions of radio galaxy classification. We show that the level of model posterior variance for individual test samples is correlated with human uncertainty when labelling radio galaxies. We explore the model performance and uncertainty calibration for different weight priors and suggest that a sparse prior produces more well-calibrated uncertainty estimates. Using the posterior distributions for individual weights, we demonstrate that we can prune 30% of the fully-connected layer weights without significant loss of performance by removing the weights with the lowest signal-to-noise ratio. A larger degree of pruning can be achieved using a Fisher information based ranking, but both pruning methods affect the uncertainty calibration for Fanaroff-Riley type I and type II radio galaxies differently. Like other work in this field, we experience a cold posterior effect, whereby the posterior must be down-weighted to achieve good predictive performance. We examine whether adapting the cost function to accommodate model misspecification can compensate for this effect, but find that it does not make a significant difference. We also examine the effect of principled data augmentation and find that this improves upon the baseline but also does not compensate for the observed effect. We interpret this as the cold posterior effect being due to the overly effective curation of our training sample leading to likelihood misspecification, and raise this as a potential issue for Bayesian deep learning approaches to radio galaxy classification in future.

astro-ph.CO↗

arXiv2021

Structured Variational Inference for Simulating Populations of Radio Galaxies

We present a model for generating postage stamp images of synthetic Fanaroff-Riley Class I and Class II radio galaxies suitable for use in simulations of future radio surveys such as those being developed for the Square Kilometre Array. This model uses a fully-connected neural network to implement structured variational inference through a variational auto-encoder and decoder architecture. In order to optimise the dimensionality of the latent space for the auto-encoder we introduce the radio morphology inception score (RAMIS), a quantitative method for assessing the quality of generated images, and discuss in detail how data pre-processing choices can affect the value of this measure. We examine the 2-dimensional latent space of the VAEs and discuss how this can be used to control the generation of synthetic populations, whilst also cautioning how it may lead to biases when used for data augmentation.

astro-ph.IM↗

arXiv2021

Fanaroff-Riley classification of radio galaxies using group-equivariant convolutional neural networks

Weight sharing in convolutional neural networks (CNNs) ensures that their feature maps will be translation-equivariant. However, although conventional convolutions are equivariant to translation, they are not equivariant to other isometries of the input image data, such as rotation and reflection. For the classification of astronomical objects such as radio galaxies, which are expected statistically to be globally orientation invariant, this lack of dihedral equivariance means that a conventional CNN must learn explicitly to classify all rotated versions of a particular type of object individually. In this work we present the first application of group-equivariant convolutional neural networks to radio galaxy classification and explore their potential for reducing intra-class variability by preserving equivariance for the Euclidean group E(2), containing translations, rotations and reflections. For the radio galaxy classification problem considered here, we find that classification performance is modestly improved by the use of both cyclic and dihedral models without additional hyper-parameter tuning, and that a D16 equivariant model provides the best test performance. We use the Monte Carlo Dropout method as a Bayesian approximation to recover epistemic uncertainty as a function of image orientation and show that E(2)-equivariant models are able to reduce variations in model confidence as a function of rotation.

astro-ph.IM↗

arXiv2021

Attention-gating for improved radio galaxy classification

In this work we introduce attention as a state of the art mechanism for classification of radio galaxies using convolutional neural networks. We present an attention-based model that performs on par with previous classifiers while using more than 50% fewer parameters than the next smallest classic CNN application in this field. We demonstrate quantitatively how the selection of normalisation and aggregation methods used in attention-gating can affect the output of individual models, and show that the resulting attention maps can be used to interpret the classification choices made by the model. We observe that the salient regions identified by the our model align well with the regions an expert human classifier would attend to make equivalent classifications. We show that while the selection of normalisation and aggregation may only minimally affect the performance of individual models, it can significantly affect the interpretability of the respective attention maps and by selecting a model which aligns well with how astronomers classify radio sources by eye, a user can employ the model in a more effective manner.

astro-ph.GA↗

arXiv2019

Cleaning radio interferometric images using a spherical wavelet decomposition

The deconvolution, or cleaning, of radio interferometric images often involves computing model visibilities from a list of clean components, in order that the contribution from the model can be subtracted from the observed visibilities. This step is normally performed using a forward fast Fourier transform (FFT), followed by a 'degridding' step that interpolates over the uv plane to construct the model visibilities. An alternative approach is to calculate the model visibilities directly by summing over all the members of the clean component list, which is a more accurate method that can also be much slower. However, if the clean components are used to construct a model image on the surface of the celestial sphere then the model visibilities can be generated directly from the wavelet coefficients, and the sparsity of the model means that most of these coefficients are zero, and can be ignored. We have constructed a prototype imager that uses a spherical-wavelet representation of the model image to generate model visibilities during each major cycle, and find empirically that the execution time scales with the wavelet resolution level, J, as O(1.07 J), and with the number of distinct clean components, N_C, as O(N_C). The prototype organises the wavelet coefficients into a tree structure, and does not store or process the zero wavelet coefficients.

astro-ph.IM↗

arXiv2019

Fast W-Projection for Wide-field Imaging

Wide-field imaging has become a major challenge for modern radio astronomy, which uses high sensitivity acquisition systems that deal with huge amounts of data. In this paper we investigate a fast wide-field imaging solution based on the w-projection algorithm, which is intended for modern astronomy systems. The core idea of the proposed method is to reduce the computational complexity of the convolution kernel generation step, specifically by replacing the standard two-dimensional FFT by the one-dimensional Hankel transform. Experimental results show that the optimised w-projection proposed here produces equivalent dirty image results in a circular image region, at a significantly lower computational cost than standard $w$-projection. One of the main advantages of the proposed solution is its slow scaling with the number of w-planes, thus enabling more accurate output results at a lower computational cost.

astro-ph.IM↗

arXiv2019

Transfer learning for radio galaxy classification

In the context of radio galaxy classification, most state-of-the-art neural network algorithms have been focused on single survey data. The question of whether these trained algorithms have cross-survey identification ability or can be adapted to develop classification networks for future surveys is still unclear. One possible solution to address this issue is transfer learning, which re-uses elements of existing machine learning models for different applications. Here we present radio galaxy classification based on a 13-layer Deep Convolutional Neural Network (DCNN) using transfer learning methods between different radio surveys. We find that our machine learning models trained from a random initialization achieve accuracies comparable to those found elsewhere in the literature. When using transfer learning methods, we find that inheriting model weights pre-trained on FIRST images can boost model performance when re-training on lower resolution NVSS data, but that inheriting pre-trained model weights from NVSS and re-training on FIRST data impairs the performance of the classifier. We consider the implication of these results in the context of future radio surveys planned for next-generation radio telescopes such as ASKAP, MeerKAT, and SKA1-MID.

astro-ph.IM↗

arXiv2018

Magnetic fields towards Ophiuchus-B derived from SCUBA-2 polarization measurements

We present the results of dust emission polarization measurements of Ophiuchus-B (Oph-B) carried out using the Submillimetre Common-User Bolometer Array 2 (SCUBA-2) camera with its associated polarimeter (POL-2) on the James Clerk Maxwell Telescope (JCMT) in Hawaii. This work is part of the B-fields In Star-forming Region Observations (BISTRO) survey initiated to understand the role of magnetic fields in star formation for nearby star-forming molecular clouds. We present a first look at the geometry and strength of magnetic fields in Oph-B. The field geometry is traced over $\sim$0.2 pc, with clear detection of both of the sub-clumps of Oph-B. The field pattern appears significantly disordered in sub-clump Oph-B1. The field geometry in Oph-B2 is more ordered, with a tendency to be along the major axis of the clump, parallel to the filamentary structure within which it lies. The degree of polarization decreases systematically towards the dense core material in the two sub-clumps. The field lines in the lower density material along the periphery are smoothly joined to the large scale magnetic fields probed by NIR polarization observations. We estimated a magnetic field strength of 630$\pm$410 $μ$G in the Oph-B2 sub-clump using a Davis-Chandeasekhar-Fermi analysis. With this magnetic field strength, we find a mass-to-flux ratio $λ$= 1.6$\pm$1.1, which suggests that the Oph-B2 clump is slightly magnetically supercritical.

astro-ph.GA↗

arXiv2018

A First Look at BISTRO Observations of The $ρ$ Oph-A core

We present 850 $μ$m imaging polarimetry data of the $ρ$ Oph-A core taken with the Submillimeter Common-User Bolometer Array-2 (SCUBA-2) and its polarimeter (POL-2), as part of our ongoing survey project, BISTRO (B-fields In STar forming RegiOns). The polarization vectors are used to identify the orientation of the magnetic field projected on the plane of the sky at a resolution of 0.01 pc. We identify 10 subregions with distinct polarization fractions and angles in the 0.2 pc $ρ$ Oph A core; some of them can be part of a coherent magnetic field structure in the $ρ$ Oph region. The results are consistent with previous observations of the brightest regions of $ρ$ Oph-A, where the degrees of polarization are at a level of a few percents, but our data reveal for the first time the magnetic field structures in the fainter regions surrounding the core where the degree of polarization is much higher ($> 5 \%$). A comparison with previous near-infrared polarimetric data shows that there are several magnetic field components which are consistent at near-infrared and submillimeter wavelengths. Using the Davis-Chandrasekhar-Fermi method, we also derive magnetic field strengths in several sub-core regions, which range from approximately 0.2 to 5 mG. We also find a correlation between the magnetic field orientations projected on the sky with the core centroid velocity components.

astro-ph.GA↗

arXiv2018

A Digital Correlator Upgrade for the Arcminute MicroKelvin Imager

The Arcminute Microkelvin Imager (AMI) telescopes located at the Mullard Radio Astronomy Observatory near Cambridge have been significantly enhanced by the implementation of a new digital correlator with 1.2 MHz spectral resolution. This system has replaced a 750-MHz resolution analogue lag-based correlator, and was designed to mitigate the effects of radio frequency interference, particularly from geostationary satellites that contaminate observations at low declinations. The upgraded instrument consists of 18 ROACH2 Field Programmable Gate Array platforms used to implement a pair of real-time FX correlators -- one for each of AMI's two arrays. The new system separates the down-converted RF baseband signal from each AMI receiver into two 2.3 GHz-wide sub-bands which are each digitized at 5-Gsps with 8 bits of precision. These digital data streams are filtered into 2048 frequency channels and cross-correlated using FPGA hardware, with a commercial 10 Gb Ethernet switch providing high-speed data interconnect. Images formed using data from the new digital correlator show over an order of magnitude improvement in dynamic range over the previous system. The ability to observe at low declinations has also been significantly improved.

astro-ph.IM↗

arXiv2017

AMI-CL J0300+2613: a Galactic anomalous-microwave-emission ring masquerading as a galaxy cluster

The Arcminute Microkelvin Imager (AMI) carried out a blind survey for galaxy clusters via their Sunyaev-Zel'dovich effect decrements between 2008 and 2011. The first detection, known as AMI-CL J0300+2613, has been reobserved with AMI equipped with a new digital correlator with high dynamic range. The combination of the new AMI data and more recent high-resolution sub-mm and infra-red maps now shows the feature in fact to be a ring of positive dust-correlated Galactic emission, which is likely to be anomalous microwave emission (AME). If so, this is the first completely blind detection of AME at arcminute scales.

astro-ph.GA↗

arXiv2017

First results from BISTRO -- a SCUBA-2 polarimeter survey of the Gould Belt

We present the first results from the B-fields In STar-forming Region Observations (BISTRO) survey, using the Sub-millimetre Common-User Bolometer Array 2 (SCUBA-2) camera, with its associated polarimeter (POL-2), on the James Clerk Maxwell Telescope (JCMT) in Hawaii. We discuss the survey's aims and objectives. We describe the rationale behind the survey, and the questions which the survey will aim to answer. The most important of these is the role of magnetic fields in the star formation process on the scale of individual filaments and cores in dense regions. We describe the data acquisition and reduction processes for POL-2, demonstrating both repeatability and consistency with previous data. We present a first-look analysis of the first results from the BISTRO survey in the OMC 1 region. We see that the magnetic field lies approximately perpendicular to the famous 'integral filament' in the densest regions of that filament. Furthermore, we see an 'hour-glass' magnetic field morphology extending beyond the densest region of the integral filament into the less-dense surrounding material, and discuss possible causes for this. We also discuss the more complex morphology seen along the Orion Bar region. We examine the morphology of the field along the lower-density north-eastern filament. We find consistency with previous theoretical models that predict magnetic fields lying parallel to low-density, non-self-gravitating filaments, and perpendicular to higher-density, self-gravitating filaments.

astro-ph.GA↗