Continual Learning on CIFAR-100 Split 10-task (test)
91.31Average Accuracy (AA)Non CL Baseline (FiLM)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Non CL Baseline (FiLM)epsilon=8, delta=1e-52024.11 | 91.31 | — | |
| Non CL Baseline (FiLM)privacy=non-DP2024.11 | 90.68 | — | |
| Non CL Baseline (last layer)epsilon=8, delta=1e-52024.11 | 88.5 | — | |
| Non CL Baseline (last layer)privacy=non-DP2024.11 | 88.4 | — | |
| Non CL Baseline (FiLM)epsilon=1, delta=1e-52024.11 | 86.06 | — | |
| PEFT Ensemble (FiLM)epsilon=8, delta=1e-52024.11 | 85.26 | 6.07 | |
| Non CL Baseline (last layer)epsilon=1, delta=1e-52024.11 | 85.1 | — | |
| PEFT Ensemble (last layer)privacy=non-DP2024.11 | 82.6 | — | |
| PEFT Ensemble (last layer)epsilon=8, delta=1e-52024.11 | 82.48 | — | |
| PEFT Ensemble (FiLM)epsilon=1, delta=1e-52024.11 | 79.79 | 7.17 | |
| PEFT Ensemble (FiLM)privacy=non-DP2024.11 | 79.39 | 10.43 | |
| Cosine classifierprivacy=non-DP2024.11 | 79.02 | 6.02 | |
| Cosine classifierepsilon=8, delta=1e-52024.11 | 78.93 | 6.15 | |
| PEFT Ensemble (last layer)epsilon=1, delta=1e-52024.11 | 78.81 | 0.06 | |
| Cosine classifierepsilon=1, delta=1e-52024.11 | 72.78 | 9.92 | |
| Naiveprivacy=non-DP2024.11 | 9.63 | — | |
| Naiveepsilon=8, delta=1e-52024.11 | 9.55 | — | |
| Naiveepsilon=1, delta=1e-52024.11 | 9.35 | 94.1 |