Table 4.

Accuracy (%) performance for binary classification. Baseline indicates that the model was fine-tuned with labeled samples only. K_c (top-N) indicates the portion of the most confident pseudolabels used. Best results are indicated in italics.

ST approach	40X	100X	200X	400X
Baseline	97.14 ± 0.33	98.22 ± 0.40	98.55 ± 0.57	98.43 ± 0.44
K c(Top-5%) pseudolabels	99.28 ± 0.6	99.44 ± 0.41	99.03 ± 0.34	99.04 ± 0.73
K c(Top-10%) pseudolabels	99.52 ± 0.33	98.85 ± 0.32	99.48 ± 0.30	99.47 ± 0.37
K c(Top-20%) pseudolabels	99.27 ± 0.37	97.95 ± 0.01	98.79 ± 0.68	98.92 ± 0.22
All pseudolabels	98.09 ± 0.21	98.20 ± 0.13	98.5 ± 0.72	98.69 ± 0.58