Image Classification on CIFAR-10 (test) (Accuracy and Retention)
1Model RetentionTeacher (ResNet-56)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Teacher (ResNet-56)Backbone=ResNet-56, Time (s)=595.2, CR=1.0x2025.12 | 1 | 0.7928 | |
| Direct TrainingTime (s)=593.7 ± 10.9, CR=3.1x2025.12 | 0.859 | — | |
| HPM-KDTime (s)=603.1 ± 52.0, CR=3.1x2025.12 | 0.854 | — | |
| TAKDTime (s)=655.5 ± 66.0, CR=3.1x2025.12 | 0.851 | — | |
| Traditional KDTime (s)=578.0 ± 6.2, CR=3.1x2025.12 | 0.847 | — | |
| Attention TransferTime (s)=588.3 ± 6.1, CR=3.1x2025.12 | 0.838 | — | |
| FitNetsTime (s)=592.4 ± 5.9, CR=3.1x2025.12 | 0.79 | — |