DeepMerge – II. Building robust deep learning algorithms for merging galaxy identification across domains

Ćiprijanović, Aleksandra; Kafkes, Diana; Downey, Kathryn; Jenkins, Sydney; Perdue, G. N.; Madireddy, Sandeep; Johnston, Travis; Snyder, Gregory F.; Nord, B.

doi:10.1093/mnras/stab1677

Cited by 31 publications

(14 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…2 The parameters for our source and target data sets are shown in Table I. Better results on all methods were achieved when starting UDA training with models that were pretrained in a supervised fashion using source data, as discussed in [67]. As such, our models were trained starting from models pre-trained on the source simulation data.…”

Section: Resultsmentioning

confidence: 99%

“…This is the same architecture that has achieved top performance in previous applications to lensing data sets [34,51,57]. More generally, CNNs are known to outperform other methods of classification for strong gravitational lenses [119], nonetheless, as noted by [67], a model trained on simulations can perform poorly on real data.…”

Section: Domain Adaptationmentioning

confidence: 85%

“…Recently, [67] used unsupervised domain adaptation (UDA) to classify merging galaxies. Doing so achieved promising results, with an increase of up to 19% in accuracy when compared to a model trained only on simulations.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Domain Adaptation for Simulation-Based Dark Matter Searches Using Strong Gravitational Lensing

Alexander¹,

Gleyzer²,

Reddy³

et al. 2021

Preprint

View full text Add to dashboard Cite

Clues to the identity of dark matter have remained surprisingly elusive, given the scope of experimental programs aimed at its identification. While terrestrial experiments may be able to nail down a model, an alternative, and equally promising, method is to identify dark matter based on astrophysical or cosmological signatures. A particularly sensitive approach is based on the unique signature of dark matter substructure on galaxy-galaxy strong lensing images. Machine learning applications have been explored in detail for extracting just this signal. With limited availability of high quality strong lensing data, these approaches have exclusively relied on simulations. Naturally, due to the differences with the real instrumental data, machine learning models trained on simulations are expected to lose accuracy when applied to real data. This is where domain adaptation can serve as a crucial bridge between simulations and real data applications. In this work, we demonstrate the power of domain adaptation techniques applied to strong gravitational lensing data with dark matter substructure. We show with simulated data sets of varying complexity, that domain adaptation can significantly mitigate the losses in the model performance. This technique can help domain experts build and apply better machine learning models for extracting useful information from strong gravitational lensing data expected from the upcoming surveys.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Domain Adaptationmentioning

confidence: 85%

See 1 more Smart Citation

Domain Adaptation for Simulation-Based Dark Matter Searches Using Strong Gravitational Lensing

Alexander¹,

Gleyzer²,

Reddy³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…It has been shown in Refs. [92,93] that domain adaptation methods offer great promise for astrophysics and cosmology. Domain adaptation methods [94], implemented during model training, help extract only features present in all datasets.…”

Section: Simulationsmentioning

confidence: 99%

Machine Learning and Cosmology

Dvorkin,

Mishra-Sharma,

Nord

et al. 2022

Preprint

View full text Add to dashboard Cite

Methods based on machine learning have recently made substantial inroads in many corners of cosmology. Through this process, new computational tools, new perspectives on data collection, model development, analysis, and discovery, as well as new communities and educational pathways have emerged. Despite rapid progress, substantial potential at the intersection of cosmology and machine learning remains untapped. In this white paper, we summarize current and ongoing developments relating to the application of machine learning within cosmology and provide a set of recommendations aimed at maximizing the scientific impact of these burgeoning tools over the coming decade through both technical development as well as the fostering of emerging communities.

show abstract

“…Presently, automated merger classification and characterization models are predominantly based on optical morphologies or other information derived from imaging (e.g. recently Pawlik et al 2016;Ackermann et al 2018;Snyder et al 2019;Walmsley et al 2019;Pearson et al 2019;Nevin et al 2019;Bottrell et al 2019b;Ćiprijanović et al 2020, 2021c. However, classification performance may be improved using other information from which a model can draw -independently, or in tandem with imaging.…”

Section: Rationale For Multi-variate Classification With Kinematicsmentioning

confidence: 99%

The combined and respective roles of imaging and stellar kinematics in identifying galaxy merger remnants

Bottrell,

Hani,

Teimoorinia

et al. 2022

Preprint

View full text Add to dashboard Cite

One of the central challenges to establishing the role of mergers in galaxy evolution is the selection of pure and complete merger samples in observations. In particular, while large and reasonably pure interacting galaxy pair samples can be obtained with relative ease via spectroscopic criteria, automated selection of post-coalescence merger remnants is restricted to the physical characteristics of remnants alone. Furthermore, such selection has predominantly focused on imaging data -whereas kinematic data may offer a complimentary basis for identifying merger remnants. Therefore, we examine the theoretical utility of both the morphological and kinematic features of merger remnants in distinguishing galaxy merger remnants from other galaxies. Deep classification models are calibrated and evaluated using idealized synthetic images and line-of-sight stellar velocity maps of a heterogeneous population of galaxies and merger remnants from the TNG100 cosmological hydrodynamical simulation. We show that even idealized stellar kinematic data has limited utility compared to imaging and under-performs by 2.1% ± 0.5% in completeness and 4.7% ± 0.4% in purity for our fiducial model architecture. Combining imaging and stellar kinematics offers a small boost in completeness (by 1.8% ± 0.4%, compared to 92.7% ± 0.2% from imaging alone) but no change in purity (0.1% ± 0.3% improvement compared to 92.7% ± 0.2%, evaluated with equal numbers of merger remnant and non-remnant control galaxies). Classification accuracy of all models is particularly sensitive to physical companions at separations 40 kpc and to time-since-coalescence. Taken together, our results show that the stellar kinematic data has little to offer in compliment to imaging for merger remnant identification in a heterogeneous galaxy population.

show abstract

DeepMerge – II. Building robust deep learning algorithms for merging galaxy identification across domains

Cited by 31 publications

References 44 publications

Domain Adaptation for Simulation-Based Dark Matter Searches Using Strong Gravitational Lensing

Domain Adaptation for Simulation-Based Dark Matter Searches Using Strong Gravitational Lensing

Machine Learning and Cosmology

The combined and respective roles of imaging and stellar kinematics in identifying galaxy merger remnants

Contact Info

Product

Resources

About