Table 7 Absolute improvement (%) wrt the SC 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 | −1.67 | 0.66 | 4.60 | 2.54 |
IS-50 | −1.06 | −1.00 | 1.83 | −0.86 |
IS-100 | −2.00 | −1.67 | 0.34 | −1.40 |
IS-250 | −1.40 | −0.60 | −0.20 | −0.33 |
GT-100 | ||||
IS-20 | −0.67 | 0.40 | 3.40 | 1.94 |
IS-50 | −2.06 | −0.06 | 1.76 | 2.74 |
IS-100 | −2.80 | −0.14 | −0.60 | 1.14 |
IS-250 | −0.33 | 0.00 | −0.27 | 0.60 |
GT-250 | ||||
IS-20 | −0.80 | 0.20 | 3.80 | 3.67 |
IS-50 | −3.06 | −0.73 | 1.56 | 0.80 |
IS-100 | −2.73 | 0.00 | −0.40 | −0.60 |
IS-250 | −2.07 | −0.60 | 0.06 | −0.80 |
GT-500 | ||||
IS-20 | −1.27 | 0.46 | 2.73 | 3.20 |
IS-50 | −1.53 | −1.20 | 0.16 | 1.60 |
IS-100 | −2.13 | −1.07 | −0.60 | −0.40 |
IS-250 | −0.87 | −0.73 | −1.34 | 0.33 |