NoisyHead

该数据集用于评估在不同隐私约束和对抗性干扰环境下，NoisyHead算法的实证行为。数据集包含了在差分隐私和对抗性设置中，测量超额测试风险和预测误差的试验。在低维设置中，数据集的规模为D=5，N=1000，并涵盖了不同的隐私级别。该数据集的任务是评估差分隐私训练方法的效能以及对抗恶意提示的鲁棒性。

This dataset is designed to evaluate the empirical performance of the NoisyHead algorithm under various privacy constraints and adversarial interference settings. It contains experiments that measure excess test risk and prediction error under both differential privacy and adversarial setups. For the low-dimensional configuration, the dataset has a feature dimension D=5 and a sample size N=1000, covering different privacy levels. The core task of this dataset is to assess the effectiveness of differentially private training methods and the robustness against malicious prompts.