How to properly set the privacy parameter in differential privacy (DP) has
been an open question in DP research since it was first proposed in 2006. In
this work, we demonstrate the ability of influence functions to offer insight
into how a specific privacy parameter value will affect a model's test loss in
the randomized response-based local DP setting. Our proposed method allows a
data curator to select the privacy parameter best aligned with their allowed
privacy-utility trade-off without requiring heavy computation such as extensive
model retraining and data privatization. We consider multiple common
randomization scenarios, such as performing randomized response over the
features, and/or over the labels, as well as the more complex case of applying
a class-dependent label noise correction method to offset the noise incurred by
randomization. Further, we provide a detailed discussion over the computational
complexity of our proposed approach inclusive of an empirical analysis. Through
empirical evaluations we show that for both binary and multi-class settings,
influence functions are able to approximate the true change in test loss that
occurs when randomized response is applied over features and/or labels with
small mean absolute error, especially in cases where noise correction methods
are applied.

我们提出了一种使用影响函数的方法来选择与允许的隐私 - 效用权衡最符合的隐私参数值，而无需进行大量的计算，例如广泛的模型训练和数据私有化。该方法适用于多种常见的随机化情景，并可以通过类别依赖的标签噪声校正方法来纠正由随机化引起的噪声。通过实证评估，我们展示了影响函数在二进制和多类别设置下，能够在特征和标签上应用随机响应时准确近似测试损失的真实变化情况，并且在应用噪声校正方法时具有较小的平均绝对误差。

通过影响函数评估本地差分隐私对效用损失的影响

Evaluating the Impact of Local Differential Privacy on Utility Loss via  Influence Functions

Arbitrary, inconsistent, or faulty decision-making raises serious concerns,
and preventing unfair models is an increasingly important challenge in Machine
Learning. Data often reflect past discriminatory behavior, and models trained
on such data may reflect bias on sensitive attributes, such as gender, race, or
age. One approach to developing fair models is to preprocess the training data
to remove the underlying biases while preserving the relevant information, for
example, by correcting biased labels. While multiple label noise correction
methods are available, the information about their behavior in identifying
discrimination is very limited. In this work, we develop an empirical
methodology to systematically evaluate the effectiveness of label noise
correction techniques in ensuring the fairness of models trained on biased
datasets. Our methodology involves manipulating the amount of label noise and
can be used with fairness benchmarks but also with standard ML datasets. We
apply the methodology to analyze six label noise correction methods according
to several fairness metrics on standard OpenML datasets. Our results suggest
that the Hybrid Label Noise Correction method achieves the best trade-off
between predictive performance and fairness. Clustering-Based Correction can
reduce discrimination the most, however, at the cost of lower predictive
performance.

通过实证方法，我们评估了六种标签噪音修正方法在多个公正度度量标准上的表现，结果表明混合标签噪音修正方法在预测性能和公正度之间取得了最佳平衡，而基于聚类的修正方法可以最大程度地减少歧视但降低了预测性能。