Table 6.
Accuracy (%) for multiclass classification.
| ST approach | 40X | 100X | 200X | 400X |
|---|---|---|---|---|
| Baseline | 91.42 ± 0.54 | 89.04 ± 0.70 | 90.07 ± 0.21 | 90.70 ± 0.63 |
| K c(top-5%) pseudolabels | 94.27 ± 0.28 | 91.78 ± 0.61 | 94.93 ± 0.17 | 92.97 ± 0.37 |
| K c(top-10%) pseudolabels | 94.28 ± 0.29 | 92.46 ± 0.48 | 94.32 ± 0.22 | 93.75 ± 0.72 |
| K c(top-20%) pseudolabels | 94.14 ± 0.14 | 93.84 ± 0.35 | 91.48 ± 0.28 | 92.19 ± 0.16 |
| All pseudolabels | 92.87 ± 0.71 | 90.41 ± 0.63 | 92.19 ± 0.38 | 91.40 ± 0.11 |
Baseline indicates that the model was fine-tuned with labeled samples only. Kc (top-N) indicates the portion of the most confident pseudolabels used. Best results are indicated in italics.