. Author manuscript; available in PMC: 2021 Nov 17.

Published in final edited form as: Neural Comput. 2021 Apr 13;33(5):1300–1328. doi: 10.1162/neco_a_01374

Table 4:

Comparison of the Training and Validation Errors of Different Algorithms for One Hidden-Layer NNs with Structured Connectivity on the MNIST Data Set.

Algorithm	Learning Rate	Training Error (%)	Validation Error (%)	Number of Epochs
R4, NPS4, Full	α_W = 0.5, 0.375; α_L =0.01	0.02	2.71	50
R4, NPS16, Full	α_W = 0.5, 0.25; α_L =0.75	0	2.41	49
R4, NPS20, Full	α_W = 0.664, 0.577; α_L =0.9	0	2.22	50
R8, NPS80, Crop	α_W = 0.664, 0.577; α_L =0.9	0.01	2.27	20
R4, NPS4, Crop	α_W = 0.099, 0.065; α_L =0.335	0.08	2.98	100
R8, NPS4, Crop	α_W = 0.099, 0.065; α_L =0.335	0	2.73	100
R8, NPS20, Crop	α_W = 0.664, 0.577; α_L =0.9	0	2.23	79