Transfer learning for galaxy morphology from one survey to another

Walker, A. R.; Zuntz, J.; Thomas, D.; Swanson, M. E. C.; Tarle, G.; Suchyta, E.; Soares-Santos, M.; Sobreira, F.; Smith, R. C.; Smith, M.; Schubnell, M.; Schindler, R.; Scarpine, V.; Sanchez, E.; Plazas, A. A.; Nord, B.; Miquel, R.; Menanteau, F.; Melchior, P.; March, M.; Maia, M. A. G.; Lahav, O.; Kuehn, K.; Kuropatkin, N.; James, D. J.; Hoyle, B.; Hollowood, D. L.; Honscheid, K.; Gutierrez, G.; Hartley, W. G.; Gschwend, J.; Gruendl, R. A.; Gruen, D.; Gerdes, D. W.; Gaztanaga, E.; García-Bellido, J.; Frieman, J.; Fosalba, P.; Evrard, A. E.; Doel, P.; De Vicente, J.; Davis, C.; da Costa, L. N.; D'Andrea, C. B.; Cunha, C. E.; Carretero, J.; Carrasco Kind, M.; Carnero Rosell, A.; Buckley-Geer, E.; Avila, S.; Brooks, D.; Annis, J.; Abbott, T. M. C.; Abdalla, F. B.; Fischer, J. L.; Kaviraj, S.; Bernardi, M.; Domínguez Sánchez, H.; Huertas-Company, M.
Bibliographical reference

Monthly Notices of the Royal Astronomical Society, Volume 484, Issue 1, p.93-100

Advertised on:
Number of authors
IAC number of authors
Refereed citations
Deep learning (DL) algorithms for morphological classification of galaxies have proven very successful, mimicking (or even improving) visual classifications. However, these algorithms rely on large training samples of labelled galaxies (typically thousands of them). A key question for using DL classifications in future Big Data surveys is how much of the knowledge acquired from an existing survey can be exported to a new data set, i.e. if the features learned by the machines are meaningful for different data. We test the performance of DL models, trained with Sloan Digital Sky Survey (SDSS) data, on Dark Energy Survey (DES) using images for a sample of ˜5000 galaxies with a similar redshift distribution to SDSS. Applying the models directly to DES data provides a reasonable global accuracy (˜90 per cent), but small completeness and purity values. A fast domain adaptation step, consisting of a further training with a small DES sample of galaxies (˜500-300), is enough for obtaining an accuracy >95 per cent and a significant improvement in the completeness and purity values. This demonstrates that, once trained with a particular data set, machines can quickly adapt to new instrument characteristics (e.g. PSF, seeing, depth), reducing by almost one order of magnitude the necessary training sample for morphological classification. Redshift evolution effects or significant depth differences are not taken into account in this study.
Related projects
Group members
Traces of Galaxy Formation: Stellar populations, Dynamics and Morphology
We are a large, diverse, and very active research group aiming to provide a comprehensive picture for the formation of galaxies in the Universe. Rooted in detailed stellar population analysis, we are constantly exploring and developing new tools and ideas to understand how galaxies came to be what we now observe.
Martín Navarro