PANOPTICON: A novel deep learning model to detect single transit events with no prior data filtering in PLATO light curves

Open Access

Table 3

Parameters and performance of every model.

N_f	Kernel			Conservative			@ 1% FAR		Label

	k_s	k_t	d	FAR (%)	Retrieval (%)	Epoch	Retrieval (%)	Epoch
4	31	11	3	0.02	81.64	56	86.23	53
	31	31	1	<0.01	67.34	66	80.58	66

	61	11	6	0.03	84.63	57	88.26	63
	61	31	2	<0.01	82.60	61	88.15	66

	121	11	12	0.37	86.45	67	89.54	52
	121	31	4	0.05	81.22	70	86.34	56

	181	11	18	0.17	86.23	56	90.07	68	A
	181	31	6	0.09	86.34	69	89.22	63

8	31	11	3	0.02	78.76	69	86.77	67
	31	31	1	0.04	82.60	54	86.66	53

	61	11	6	<0.01	85.81	64	89.64	53	B, C
	61	31	2	0.02	84.20	57	89.43	66

	121	11	12	0.09	86.65	58	89.54	61
	121	31	4	0.02	84.95	63	88.26	64

	181	11	18	0.17	87.19	63	89.43	69
	181	31	6	0.03	84.52	52	89.43	69

Notes. Performance of the models tested for this study. We considered two approaches to evaluate the results of the models and compared them for each initial number of features N_f, kernel size k_s, and trainable parameters in the kernel k_t. The conservative approach presents the model with the lowest FAR possible, and the second approach consists of fixing the FAR at a value of 1% and taking the highest resulting recovery. Models B and C come from the same training round and thus share the same kernel parameters, but they stem from two different epochs.

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.