A deployed real-time end-to-end deep learning algorithm for fast radio burst detection

Open Access

Table 2

Average testing accuracy sampled over three independently seeded models per model architecture, recall rate, FPR, and F1 scores for all benchmarked models and their variants.

Model	Acc. %	Rec. %	FP %	F1
SPANDAK CNN	92.7	98.0	12.6	0.930
Mask ResNet	99.0±0.2	99.1±0.1	1.18±0.45	0.990±0.001
8bit Mask ResNet	99.0±0.1	98.6±0.1	0.60±0.20	0.990±0.001
Thres. Mask ResNet	99.1±0.1	98.8±0.2	0.49±0.09	0.991±0.001
Vanilla ResNet	98.2±0.7	98.1±0.1	1.93±1.32	0.981±0.068

Notes. The uncertainty represents 1 σ deviation over the individual seeded results. All models were seeded except for SPANDAK CNN as only one seeded model was made available. Boldfaced results indicate best performing in each respective category.

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.