1 paper
Hao Li, Zheng Li, Siyuan Wu +5
Most existing membership inference attacks (MIAs) utilize metrics (e.g., loss) calculated on the model's final state, while recent advanced attacks leverage metrics computed at var…