Optimized Tradeoffs for Private Prediction with Majority Ensembling