TABLE 1.

Area under curve (AUC) mean and standard deviation values of multi-scale, domain knowledge-guided attention (MSGA) performance on validation set for various λ^h and λ^m (task importance) parameters

		λm
		0	1	10	50	100	200
λh	200	0.87 (0.14)	0.98 (0.02)	0.88 (0.21)	0.89 (0.18)	0.87 (0.21)	0.97 (0.02)
	100	0.85 (0.20)	0.96 (0.04)	0.86 (0.20)	0.90 (0.10)	0.84 (0.21)	0.97 (0.03)
	50	0.83 (0.20)	0.88 (0.09)	0.89 (0.22)	0.84 (0.22)	0.97 (0.01)	0.98 (0.02)
	10	0.87 (0.21)	0.92 (0.09)	0.84 (0.17)	0.85 (0.21)	0.99 (0.01)	0.81 (0.23)
	1	0.87 (0.18)	0.84 (0.21)	0.95 (0.07)	0.89 (0.08)	0.89 (0.12)	0.76 (0.23)
	0	0.93 (0.07)	0.93 (0.07)	0.93 (0.09)	0.86 (0.15)	0.94 (0.04)	0.85 (0.21)

Note: λ^h and λ^m are the relative task importance parameters in the overall loss function, representing high- and medium-resolution attentions, respectively. Three top performing combinations (λ^h = 200 and λ^m = 1; λ^h = 50 and λ^m = 200; λ^h = 10 and λ^m = 100) are in bold font.