Node Classification on Pubmed (test) (Unlearned Performance)
0.02Accuracy Drop (ΔAcc)TDGIA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TDGIAr=0.001, nv=19, Defending Model=GCN2026.03 | 0.02 | — | — | — | 0.03 | |
| ATDGIAr=0.001, nv=19, Defending Model=GCN2026.03 | 0.03 | — | — | — | 0.02 | |
| AGIAr=0.001, nv=19, Defending Model=GCN2026.03 | 0.04 | — | — | — | 0.03 | |
| PEAr=0.001, nv=19, Defending Model=GCN2026.03 | 0.21 | — | — | — | 0.33 | |
| TDGIAr=0.01, nv=197, Defending Model=GCN2026.03 | 0.72 | — | — | — | 0.72 | |
| AGIAr=0.01, nv=197, Defending Model=GCN2026.03 | 0.76 | — | — | — | 0.76 | |
| ATDGIAr=0.01, nv=197, Defending Model=GCN2026.03 | 1.1 | — | — | — | 1.16 | |
| TDGIAr=0.05, nv=985, Defending Model=GCN2026.03 | 3.78 | — | — | — | 3.88 | |
| AGIAr=0.05, nv=985, Defending Model=GCN2026.03 | 3.99 | — | — | — | 4.14 | |
| PEAr=0.01, nv=197, Defending Model=GCN2026.03 | 4.33 | — | — | — | 7.24 | |
| ATDGIAr=0.05, nv=985, Defending Model=GCN2026.03 | 10.22 | — | — | — | 10.65 | |
| TDGIAr=0.1, nv=1971, Defending Model=GCN2026.03 | 11.54 | — | — | — | 11.88 | |
| AGIAr=0.1, nv=1971, Defending Model=GCN2026.03 | 11.98 | — | — | — | 12.39 | |
| ATDGIAr=0.1, nv=1971, Defending Model=GCN2026.03 | 11.98 | — | — | — | 12.29 | |
| OptimAttackUnlearn Method=GA2026.03 | 14.81 | 82.79 | 67.98 | — | — | |
| PEAr=0.05, nv=985, Defending Model=GCN2026.03 | 34.24 | — | — | — | 50.24 | |
| PEAr=0.1, nv=1971, Defending Model=GCN2026.03 | 41.78 | — | — | — | 61.09 | |
| OptimAttackUnlearn Method=GIF2026.03 | 59.08 | 83.51 | 24.43 | — | — | |
| OptimAttackUnlearn Method=CEU2026.03 | 59.56 | 81.02 | 21.46 | — | — | |
| Clean GraphDefending Model=GCN2026.03 | — | 85.8 | — | 84.71 | — |