Table 2. Average computation time (in seconds, measured for 100 simulated networks) required to train the ANN for different values of K ratio, learning rate and number of iteration of gradient descent.

Ntrain	LR = 10−3			LR = 10−4			LR = 10−5
	100	1,000	2,000	100	1,000	2,000	100	1,000	2,000
K = 20	12.08	107.7	213.66	12	107.7	214.36	11.91	107.8	213.72
K = 10	7.6	72.8	145.1	7.68	72.8	145.1	7.61	72.88	145.28
K = 3	3.4	33.12	65.9	3.44	33.25	66.1	3.4	33.18	66.22
K = 1	2.6	25.9	51.7	2.64	25.98	51.69	2.6	26	51.82