Table 7 Explainability performance comparison on the ISIC 2019 dataset.
From: MTAKD: multi-teacher agreement knowledge distillation for edge AI skin disease diagnosis
Framework | Insertion AUC | Deletion AUC | Insertion \(\:\varvec{p}\)-value vs. MTAKD | Deletion \(\:\varvec{p}\)-value vs. MTAKD |
|---|---|---|---|---|
Hinton-KD 8 | 0.6155 ± 0.0194 | 0.4526 ± 0.0099 | 0.0031 | 0.02352 |
AT-KD 20 | 0.6020 ± 0.0451 | 0.4710 ± 0.0369 | 0.0162 | 0.04438 |
RKD 16 | 0.5749 ± 0.0590 | 0.5288 ± 0.0436 | 0.0149 | 0.00220 |
SSD-KD 10 | 0.5199 ± 0.0404 | 0.5422 ± 0.0362 | 0.0010 | 0.00366 |
Averaging-KD 12 | 0.6191 ± 0.0230 | 0.4640 ± 0.0387 | 0.0044 | 0.03530 |
CA-MKD 13 | 0.5948 ± 0.0168 | 0.4639 ± 0.0400 | 0.0005 | 0.03946 |
MTAKD | 0.6796 ± 0.0044 | 0.4090 ± 0.0198 | - | - |
