Light-curve classification with recurrent neural networks for GOTO: dealing with imbalanced data

Burhanudin, U.; Maund, Justyn R.; Killestein, T.; Ackley, K.; Dyer, M.; Lyman, J.; Ulaczyk, K.; Cutter, R.; Mong, Y. L.; Steeghs, D.; Galloway, D. K.; Dhillon, V. S.; O’Brien, P. T.; Ramsay, Gavin; Noysena, K.; Kotak, R.; Breton, R. P.; Nuttall, L. K.; Πάλλη, Ε.; Pollacco, D.; Thrane, E.; Awiphan, S.; Chote, P.; Chrimes, A. A.; Daw, E. J.; Duffy, C.; Eyles-Ferris, R. A. J.; Gompertz, B. P.; Heikkilä, T.; Irawati, P.; Kennedy, Mark; Levan, A. J.; Littlefair, S.; Makrygianni, L.; Sánchez, D. Mata; Mattila, S.; McCormac, J.; Mkrtichian, D. E.; Mullaney, James; Sawangwit, U.; Stanway, Elizabeth R.; Starling, R. L. C.; Strøm, Paul A.; Tooke, S.; Wiersema, K.

doi:10.1093/mnras/stab1545

Cited by 23 publications

(16 citation statements)

References 67 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recurrent neural networks have also been developed to provide source classification based on their light-curves. 16 Work is ongoing to make a publicly-accessible version of the Marshall, utilising the successful Zooniverse platform. The GOTO prototype was deployed at the Observatorio del Roque de los Muchachos on La Palma in the Canary Islands in the summer of 2017.…”

Section: Source Identificationmentioning

confidence: 99%

The Gravitational-wave Optical Transient Observer (GOTO)

Dyer

Ackley

Lyman

et al. 2022

Ground-Based and Airborne Telescopes IX

View full text Add to dashboard Cite

The Gravitational-wave Optical Transient Observer (GOTO) is a wide-field telescope project focused on detecting optical counterparts to gravitational-wave sources. Each GOTO robotic mount holds eight 40 cm telescopes, giving an overall field of view of 40 square degrees. As of 2022 the first two GOTO mounts have been commissioned at the Roque de los Muchachos Observatory on La Palma, Canary Islands, and construction of the second node with two additional 8-telescope mounts has begin at Siding Spring Observatory in New South Wales, Australia. Once fully operational each GOTO mount will be networked to form a robotic, multi-site observatory, which will survey the entire visible sky every two nights and enable rapid follow-up detections of transient sources.

show abstract

Section: Source Identificationmentioning

confidence: 99%

The Gravitational-wave Optical Transient Observer (GOTO)

Dyer

Ackley

Lyman

et al. 2022

Ground-Based and Airborne Telescopes IX

View full text Add to dashboard Cite

show abstract

“…Learning from an imbalanced dataset can be difficult, since conventional algorithms assume an even distribution of classes within the data set. Classifiers will tend to misclassify examples from the minority class, and will be optimised to perform well on classifying examples from the majority class (see Burhanudin et al 2021).…”

Section: Selection Cutsmentioning

confidence: 99%

“…Takahashi et al (2020) used a neural network to classify supernova light curves from the Hyper Suprime-Cam transient survey. In Burhanudin et al (2021), we presented a recurrent neural network for classifying light curves from the Gravitational-wave Optical Transient Observer (GOTO) survey, capable of handling an imbalanced training dataset. In all the examples listed above, the overall classification performance is good for a number of classification tasks (binary Ia/non-Ia or a multi-class problem into the different supernova subtypes), achieving accuracies of 85%.…”

Section: Introductionmentioning

confidence: 99%

Pan-chromatic photometric classification of supernovae from multiple surveys and transfer learning for future surveys

Burhanudin¹,

Maund²

2022

Preprint

View full text Add to dashboard Cite

Time-domain astronomy is entering a new era as wide-field surveys with higher cadences allow for more discoveries than ever before. The field has seen an increased use of machine learning and deep learning for automated classification of transients into established taxonomies. Training such classifiers requires a large enough and representative training set, which is not guaranteed for new future surveys such as the Vera Rubin Observatory, especially at the beginning of operations. We present the use of Gaussian processes to create a uniform representation of supernova light curves from multiple surveys, obtained through the Open Supernova Catalog for supervised classification with convolutional neural networks. We also investigate the use of transfer learning to classify light curves from the Photometric LSST Astronomical Time Series Classification Challenge (PLAsTiCC) dataset. Using convolutional neural networks to classify the Gaussian process generated representation of supernova light curves from multiple surveys, we achieve an AUC score of 0.859 for classification into Type Ia, Ibc, and II. We find that transfer learning improves the classification accuracy for the most under-represented classes by up to 18% when classifying PLAsTiCC light curves, and is able to achieve an AUC score of 0.945 when including photometric redshifts for classification into six classes (Ia, Iax, Ia-91bg, Ibc, II, SLSN-I). We also investigate the usefulness of transfer learning when there is a limited labelled training set to see how this approach can be used for training classifiers in future surveys at the beginning of operations.

show abstract

“…[21,22]), morphological classification (e.g., Refs. [23][24][25][26][27]), source selection and classification (e.g., Refs. [28][29][30][31]), image and spectral reconstruction (e.g., Ref.…”

Section: Introductionmentioning

confidence: 99%

Exploring New Redshift Indicators for Radio-Powerful AGN

et al. 2021

View full text Add to dashboard Cite

Active Galactic Nuclei (AGN) are relevant sources of radiation that might have helped reionising the Universe during its early epochs. The super-massive black holes (SMBHs) they host helped accreting material and emitting large amounts of energy into the medium. Recent studies have shown that, for epochs earlier than z∼5, the number density of SMBHs is on the order of few hundreds per square degree. Latest observations place this value below 300 SMBHs at z≳6 for the full sky. To overcome this gap, it is necessary to detect large numbers of sources at the earliest epochs. Given the large areas needed to detect such quantities, using traditional redshift determination techniques—spectroscopic and photometric redshift—is no longer an efficient task. Machine Learning (ML) might help obtaining precise redshift for large samples in a fraction of the time used by other methods. We have developed and implemented an ML model which can predict redshift values for WISE-detected AGN in the HETDEX Spring Field. We obtained a median prediction error of σzN=1.48×(zPredicted−zTrue)/(1+zTrue)=0.1162 and an outlier fraction of η=11.58% at (zPredicted−zTrue)/(1+zTrue)>0.15, in line with previous applications of ML to AGN. We also applied the model to data from the Stripe 82 area obtaining a prediction error of σzN=0.2501.

show abstract

Light-curve classification with recurrent neural networks for GOTO: dealing with imbalanced data

Cited by 23 publications

References 67 publications

The Gravitational-wave Optical Transient Observer (GOTO)

The Gravitational-wave Optical Transient Observer (GOTO)

Pan-chromatic photometric classification of supernovae from multiple surveys and transfer learning for future surveys

Exploring New Redshift Indicators for Radio-Powerful AGN

Contact Info

Product

Resources

About