Membership Inference Attack on Twitter (TPR vs FPR evaluation)
13.9TPR @ 1% FPRLikelihood Ratio Attack (Oracle Reference Model*)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Likelihood Ratio Attack (Oracle Reference Model*)Attack Category=Likelihood Ratio Attack, Reference Source=Oracle2023.05 | 13.9 | 1.59 | 0.28 | |
| Neighbour AttackAttack Category=Reference-Free Attack2023.05 | 7.35 | 1.43 | 0.28 | |
| Likelihood Ratio Attack (Candidate Reference Model 2)Attack Category=Likelihood Ratio Attack, Reference Source=offensive tweet classification dataset2023.05 | 6.61 | 1.19 | 0.25 | |
| Likelihood Ratio Attack (Candidate Reference Model 1)Attack Category=Likelihood Ratio Attack, Reference Source=Twitter mental health dataset2023.05 | 6.49 | 1.1 | 0.24 | |
| Likelihood Ratio Attack (Base Reference Model)Attack Category=Likelihood Ratio Attack, Reference Source=Base Reference Model2023.05 | 5.66 | 0.98 | 0.22 | |
| LOSS AttackAttack Category=Reference-Free Attack2023.05 | 2.08 | 0.11 | 0.02 |