English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Machine learning is becoming a popular tool to quantify galaxy morphologiesand identify mergers. However, this technique relies on using an appropriateset of training data to be successful. By combining hydrodynamical simulations,synthetic observations and convolutional neural networks (CNNs), wequantitatively assess how realistic simulated galaxy images must be in order toreliably classify mergers. Specifically, we compare the performance of CNNstrained with two types of galaxy images, stellar maps and dust-inclusiveradiatively transferred images, each with three levels of observationalrealism: (1) no observational effects (idealized images), (2) realistic sky andpoint spread function (semi-realistic images), (3) insertion into a real skyimage (fully realistic images). We find that networks trained on eitheridealized or semi-real images have poor performance when applied tosurvey-realistic images. In contrast, networks trained on fully realisticimages achieve 87.1% classification performance. Importantly, the level ofrealism in the training images is much more important than whether the imagesincluded radiative transfer, or simply used the stellar maps (87.1% compared to79.6% accuracy, respectively). Therefore, one can avoid the large computationaland storage cost of running radiative transfer with a relatively modestcompromise in classification performance. Making photometry-based networksinsensitive to colour incurs a very mild penalty to performance withsurvey-realistic data (86.0% with r-only compared to 87.1% with gri). Thisresult demonstrates that while colour can be exploited by colour-sensitivenetworks, it is not necessary to achieve high accuracy and so can be avoided ifdesired. We provide the public release of our statistical observational realismsuite, RealSim, as a companion to this paper.

Deep learning predictions of galaxy merger stage and the importance of observational realism