Deep learning models are increasingly popular in many machine learning applications where the training data may contain sensitive information. To provide formal and rigorous privacy guarantee, many learning systems now incorporate differential privacy by training their models with (differentially) private SGD. A key step in each private SGD update is gradient clipping that shrinks the gradient of an individual example whenever its L2 norm exceeds some threshold. We first demonstrate how gradient clipping can prevent SGD from converging to stationary point. We then provide a theoretical analysis that fully quantifies the clipping bias on convergence with a disparity measure between the gradient distribution and a geometrically symmetric distribution. Our empirical evaluation further suggests that the gradient distributions along the trajectory of private SGD indeed exhibit symmetric structure that favors convergence. Together, our results provide an explanation why private SGD with gradient clipping remains effective in practice despite its potential clipping bias. Finally, we develop a new perturbation-based technique that can provably correct the clipping bias even for instances with highly asymmetric gradient distributions.
翻译:为了提供正式和严格的隐私保障,许多学习系统现在都通过(不同地)用私人 SGD 来培训自己的模型,从而纳入不同的隐私。每个私人 SGD 更新的关键步骤是梯度剪切,当个人例的L2 规范超过某些阈值时,这种剪切会缩缩缩梯度。我们首先演示梯度剪切如何防止SGD 凝聚到固定点。我们然后提供理论分析,充分量化关于渐变分布和几何对称分布之间差异测量的切合偏差。我们的经验评估进一步表明,私人 SGD 轨迹上的梯度分布确实展示了有利于趋同的对称结构。我们的结果共同解释了为什么带有梯度剪切的私人SGD 切除在实际中仍然有效,尽管它具有潜在的偏差偏差。最后,我们开发了一种新的以扰动为基础的技术,可以证明即使在高度不对称的梯度分布的情况下也能纠正剪切偏差的偏差。