Machine Unlearning on Tiny-ImageNet (test)
99.65Residual AccuracyRT
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| RTBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 99.65 | — | 57.81 | — | 54.1 | — | — | 46.72 | 0 | 33.73 | |
| FTBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 99.44 | — | 15.85 | — | 56.53 | — | — | 5.44 | 21.47 | 4.2 | |
| NegTVBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 98.81 | — | 6.95 | — | 56.04 | — | — | 1.85 | 24.63 | 0.97 | |
| RLBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 98.77 | — | 83.52 | — | 52.61 | — | — | 30.49 | 11.08 | 14.11 | |
| SFRonBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 97.94 | — | 23.2 | — | 52.27 | — | — | 33.12 | 12.94 | 20.56 | |
| MCUBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 97.73 | — | 47.25 | — | 52.35 | — | — | 38.38 | 5.64 | 9.78 | |
| SalUnBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 97.66 | — | 86.07 | — | 53.32 | — | — | 39.55 | 9.55 | 13.73 | |
| MCUβBackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 96.94 | — | 45.25 | — | 50.75 | — | — | 44.72 | 5.16 | 8.44 | |
| GABackbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 95.91 | — | 7.83 | — | 53.6 | — | — | 4.42 | 24.13 | 0.5 | |
| LOFTD_rm=only Σ_rm^pre, D_fg=only Σ_fg^pre, Backbone=Swin-T, RTE=0.29, # param. (%)=13,824 (0.05%)2026.01 | 92.17 | 0.02 | 0.08 | 0.42 | — | — | — | — | — | — | |
| FTD_rm=true, D_fg=false, Backbone=Swin-T, RTE=47.42, # param. (%)=27,673,154 (100%)2026.01 | 91.13 | 49.98 | 38.82 | 31.88 | — | — | — | — | — | — | |
| retrainedD_rm=true, D_fg=false, Backbone=Swin-T, RTE=95.17, # param. (%)=27,673,154 (100%)2026.01 | 90.57 | 0 | 0 | 0 | — | — | — | — | — | — | |
| RLD_rm=true, D_fg=true, Backbone=Swin-T, RTE=431.45, # param. (%)=27,673,154 (100%)2026.01 | 90.03 | 15.38 | 1.1 | 7.72 | — | — | — | — | — | — | |
| SalUnD_rm=true, D_fg=true, Backbone=Swin-T, RTE=476.78, # param. (%)=13,836,577 (50%)2026.01 | 87.93 | 11.61 | 2.74 | 6.31 | — | — | — | — | — | — | |
| NegGrad+Backbone=VGG-16-BN, Forgetting Ratio=20%2025.05 | 85.23 | — | 40.13 | — | 47.55 | — | — | 45.02 | 10.09 | 4.22 | |
| DELETED_rm=false, D_fg=true, Backbone=Swin-T, RTE=446.62, # param. (%)=27,673,154 (100%)2026.01 | 81.07 | 10.96 | 1.16 | 6.23 | — | — | — | — | — | — | |
| pretrainedD_rm=true, D_fg=true, Backbone=Swin-T, RTE=882.05, # param. (%)=27,673,154 (100%)2026.01 | 75.43 | 74.45 | 95.34 | — | — | — | — | — | — | — | |
| COUN+RLD_rm=false, D_fg=true, Backbone=Swin-T, RTE=337.69, # param. (%)=27,673,154 (100%)2026.01 | 33.1 | 24.07 | 3.28 | 35.27 | — | — | — | — | — | — | |
| RLD_rm=false, D_fg=true, Backbone=Swin-T, RTE=195.26, # param. (%)=27,673,154 (100%)2026.01 | 23.43 | 13.75 | 5.39 | 35.12 | — | — | — | — | — | — | |
| SalUnD_rm=false, D_fg=true, Backbone=Swin-T, RTE=235.23, # param. (%)=13,836,577 (50%)2026.01 | 23.3 | 15.05 | 12.25 | 36.95 | — | — | — | — | — | — | |
| BTD_rm=false, D_fg=true, Backbone=Swin-T, RTE=88.81, # param. (%)=27,673,154 (100%)2026.01 | 21.3 | 12.57 | 4.09 | 35.16 | — | — | — | — | — | — | |
| L2ULD_rm=false, D_fg=true, Backbone=Swin-T, RTE=7361.76, # param. (%)=27,673,154 (100%)2026.01 | 19.1 | 11.91 | 3.66 | 35.63 | — | — | — | — | — | — | |
| GAD_rm=false, D_fg=true, Backbone=Swin-T, RTE=178.07, # param. (%)=27,673,154 (100%)2026.01 | 12.2 | 8.1 | 11.41 | 38.64 | — | — | — | — | — | — | |
| DELETEBackbone=ViT2026.03 | — | — | — | — | 0 | 37.33 | 89.64 | — | — | — | |
| FTBackbone=ViT2026.03 | — | — | — | — | 88.33 | 93.65 | 89.19 | — | — | — | |
| GABackbone=ViT2026.03 | — | — | — | — | 1.56 | 16.57 | 75.85 | — | — | — | |
| GDRBackbone=ViT2026.03 | — | — | — | — | 24.22 | 85.14 | 85.42 | — | — | — | |
| ℓ1-sparseBackbone=ViT2026.03 | — | — | — | — | 78.11 | 89.27 | 78.91 | — | — | — | |
| MunbaBackbone=ViT2026.03 | — | — | — | — | 75.57 | 90.41 | 83.04 | — | — | — | |
| OriginalBackbone=ViT2026.03 | — | — | — | — | 89.38 | 94.95 | 93.33 | — | — | — | |
| Our methodBackbone=ViT2026.03 | — | — | — | — | 3.11 | 91.27 | 88.88 | — | — | — | |
| Retrain2026.03 | — | — | — | — | 64.45 | 95.71 | 90.58 | — | — | — | |
| SalUnBackbone=ViT2026.03 | — | — | — | — | 5.67 | 55.81 | 73 | — | — | — | |
| SCRUBBackbone=ViT2026.03 | — | — | — | — | 7.22 | 81.97 | 88.29 | — | — | — | |
| SSDBackbone=ViT2026.03 | — | — | — | — | 44.33 | 77.05 | 87.9 | — | — | — | |
| Two-Stage Machine Unlearning2026.03 | — | — | — | — | 3.11 | 91.27 | 88.88 | — | — | — |