HOLISMOKES - XI. Evaluation of supervised neural networks for strong-lens searches in ground-based imaging surveys

Open Access

Table 5

Performance for various fractions of the overall data set.

	CNN			ResNet

Fraction	AUROC	TPR₀	TPR₁₀	AUROC	TPR₀	TPR₁₀
10%	0.9806	0.0	41.3	0.9614	5.8	38.1
20%	0.9582	0.0	51.9	0.9532	3.7	16.9
30%	0.9467	0.0	28.0	0.9675	12.7	36.0
40%	0.9426	0.0	33.9	0.9063	16.4	27.0
50%	0.9530	0.0	38.6	0.8583	6.4	22.2
60%	0.9661	0.0	42.3	0.9568	15.3	24.3
70%	0.9623	0.0	45.5	0.9637	21.2	37.6
80%	0.9592	0.0	44.9	0.9686	25.9	45.0
90%	0.9483	5.8	31.8	0.9820	8.5	44.4
100%	0.9557	0.0	44.4	0.9913	36.0	55.0

Notes. The baseline training set is used together with the baseline CNN and ResNet.

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.