Deep convolutional neural networks as strong gravitational lens detectors

C. Schaefer; M. Geiger; T. Kuntzer; J.-P. Kneib

doi:10.1051/0004-6361/201731201

Home

All issues

Volume 611 (March 2018)

A&A, 611 (2018) A2

Full HTML

Free Access

Table 1

Training results: rach architecture was run five separate times.

Space	Training	Validation
baseline	0.9920 ± 0.0020	0.9764 ± 0.0017
views	0.9898 ± 0.0030	0.9753 ± 0.0021
residual	0.9958 ± 0.0028	0.9765 ± 0.0024
invariant	0.9997 ± 0.0003	0.9719 ± 0.0016

Ground	Training	Validation

baseline	0.9953 ± 0.0090	0.9905 ± 0.0082
views	0.9980 ± 0.0010	0.9924 ± 0.0006
residual	0.9990 ± 0.0023	0.9932 ± 0.0027
invariant	0.9999 ± 3 × 10⁻⁶	0.9880 ± 0.0030

Notes. The training and validation AUC scores are the mean of these runs. The error is the standard deviation.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.