Skip to main content

Table 5 Absolute improvement (%) wrt the NMF baseline when bases are learned from the unlabeled GT data sets

From: High level feature extraction for the self-taught learning algorithm

Training Dictionary size K
Set 100 200 300 500
GT-50
IS-20 −1.53 0.27 −0.93 −0.54
IS-50 −0.47 1.13 1.20 1.33
IS-100 0.40 −0.07 −0.33 0.33
IS-250 −1.07 0.00 0.20 0.00
GT-100
IS-20 −1.06 −0.2 0.07 −0.80
IS-50 −0.67 1.20 1.67 1.20
IS-100 0.40 −0.40 −0.33 0.46
IS-250 −0.20 −0.33 0.67 0.00
GT-250
IS-20 −2.53 −0.26 0.47 −0.34
IS-50 −0.93 1.60 1.67 1.13
IS-100 −0.06 0.27 0.47 −0.80
IS-250 −0.27 0.00 0.07 0.00
GT-500
IS-20 −3.06 −0.93 0.27 −0.94
IS-50 −1.67 1.87 1.33 1.06
IS-100 −0.40 −0.33 −0.13 −0.74
IS-250 −0.87 −0.40 −0.67 0.00