Continual Unlearning on CIFAR-100 ResNet-20 (test)
60.46Retention AccuracyMoE Router
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MoE RouterTasks (N)=20, lambda=0.52026.04 | 60.46 | 55 | |
| MoE RouterTasks (N)=20, lambda=1.02026.04 | 60.46 | 55 | |
| MoE RouterTasks (N)=5, lambda=0.52026.04 | 59.8 | 51.2 | |
| MoE RouterTasks (N)=5, lambda=1.02026.04 | 59.8 | 51.2 | |
| MoE RouterTasks (N)=10, lambda=0.52026.04 | 59.62 | 57 | |
| MoE RouterTasks (N)=10, lambda=1.02026.04 | 59.62 | 57 | |
| MoE RouterTasks (N)=30, lambda=0.52026.04 | 59.49 | 59.1 | |
| MoE RouterTasks (N)=30, lambda=1.02026.04 | 59.49 | 59.1 | |
| In-Training ApproachTasks (N)=20, lambda=0.52026.04 | 59.39 | 50.75 | |
| In-Training ApproachTasks (N)=5, lambda=0.52026.04 | 59.04 | 37.6 | |
| In-Training ApproachTasks (N)=30, lambda=0.52026.04 | 58.76 | 56.13 | |
| In-Training ApproachTasks (N)=10, lambda=0.52026.04 | 58.68 | 50.7 | |
| In-Training ApproachTasks (N)=20, lambda=1.02026.04 | 58.23 | 48 | |
| In-Training ApproachTasks (N)=30, lambda=1.02026.04 | 58.06 | 53.83 | |
| In-Training ApproachTasks (N)=5, lambda=1.02026.04 | 58.02 | 33.8 | |
| In-Training ApproachTasks (N)=10, lambda=1.02026.04 | 57.19 | 48.4 | |
| Static / AC-LoRA, I-LoRATasks (N)=5, lambda=0.52026.04 | 56.23 | 5.6 | |
| Static / AC-LoRA, I-LoRATasks (N)=10, lambda=0.52026.04 | 51.89 | 13.8 | |
| Static / AC-LoRA, I-LoRATasks (N)=5, lambda=1.02026.04 | 48.78 | 0 | |
| Static / AC-LoRA, I-LoRATasks (N)=20, lambda=0.52026.04 | 45.3 | 10.35 | |
| Static / AC-LoRA, I-LoRATasks (N)=10, lambda=1.02026.04 | 37.58 | 1.2 | |
| Static / AC-LoRA, I-LoRATasks (N)=30, lambda=0.52026.04 | 35.21 | 10.6 | |
| Static / AC-LoRA, I-LoRATasks (N)=20, lambda=1.02026.04 | 25.19 | 0.25 | |
| Static / AC-LoRA, I-LoRATasks (N)=30, lambda=1.02026.04 | 12.74 | 0.77 |