针对医疗图像数据不平衡问题，提出自适应降噪梯度方法提升隐私保护下模型性能。
Steps Adaptive Decay DPSGD: Enhancing Performance on Imbalanced Datasets with Differential Privacy with HAM10000
- 用线性衰减机制动态调整噪声与裁剪阈值，初期保留更多少数类信息。
- 在HAM10000数据集上，ε=3.0、δ=10⁻³时准确率提升2.15%。
- 适合小样本、类别不平衡的医疗图像隐私学习场景。
在医疗图像分类中，数据泄露是关键问题。以往通过向梯度添加噪声实现差分隐私的方法在大型数据集（如MNIST、CIFAR-100）上表现良好，但在小型、类别不平衡的医疗数据集（如HAM10000）上失效。这是由于不平衡分布导致少数类的梯度被过度裁剪而丢失关键信息，多数类则占据主导地位，使模型过早陷入次优解。为此，我们提出SAD-DPSGD，采用线性衰减机制调节噪声和裁剪阈值。通过在训练初期分配更多隐私预算并使用更高裁剪阈值，模型得以避免次优解，提升性能。实验表明，SAD-DPSGD在HAM10000上优于Auto-DPSGD，于ε=3.0、δ=10⁻³条件下准确率提升2.15%。
原文摘要 · Abstract (English)
When applying machine learning to medical image classification, data leakage is a critical issue. Previous methods, such as adding noise to gradients for differential privacy, work well on large datasets like MNIST and CIFAR-100, but fail on small, imbalanced medical datasets like HAM10000. This is because the imbalanced distribution causes gradients from minority classes to be clipped and lose crucial information, while majority classes dominate. This leads the model to fall into suboptimal solutions early. To address this, we propose SAD-DPSGD, which uses a linear decaying mechanism for noise and clipping thresholds. By allocating more privacy budget and using higher clipping thresholds in the initial training phases, the model avoids suboptimal solutions and enhances performance. Experiments show that SAD-DPSGD outperforms Auto-DPSGD on HAM10000, improving accuracy by 2.15% under $ε= 3.0$ , $δ= 10^{-3}$.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。