Membership Inference Attack on Blog Authorship Corpus (test)
86.4AccuracyPRIME
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PRIMEAttack Configuration=All three tasks (3-way), Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 86.4 | 86 | 88.8 | |
| PRIMEAttack Configuration=Horoscope + Age (2-way), Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 84.9 | 84.4 | 87.3 | |
| PRIMEAttack Configuration=Gender + Horoscope (2-way), Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 83.5 | 83.1 | 85.9 | |
| Single-Task MIAAttack Configuration=Horoscope, Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 78.7 | 77.9 | 83.2 | |
| PRIMEAttack Configuration=Gender + Age (2-way), Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 76.4 | 75.8 | 79.2 | |
| Single-Task MIAAttack Configuration=Age, Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 66.4 | 65.1 | 73 | |
| Single-Task MIAAttack Configuration=Gender, Base Model=Qwen3-1.7B, Fine-tuning=LoRA2026.07 | 61.8 | 60.2 | 66.3 |