Similarity-based pairing improves efficiency of siamese neural networks for regression tasks and uncertainty quantification

Table 1 Overview of hyperparameters used

Model	Structure	Max LR	Min LR	Patience	Epsilon	Epochs
MLP	[2048,128,1]	5e-4	1e-5	50	1e-6	70
MLP-ΔFP^a	[2048,128,1]	1e-3	1e-5	50	1e-6	200/150/100
MLP-ΔFP^b	[2048,128,1]	1e-4	1e-6	200	1e-7	100
MLP-SNN^a	[2048,512,128;128,1]	1e-3	1e-5	50	1e-6	200,150,100
MLP-SNN^b	[2048,512,128;128,1]	1e-4	1e-6	200	1e-7	20/40/30
Chemformer	[Dimension = 512, Attention Heads = 8, Layers = 6] [512,64,1]	5e-4	5e-4	0	1e-8	150

ISSN: 1758-2946