arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2106.09352cs.LGcs.CR
收藏

基于低秩重参数化的大规模私人学习

Large Scale Private Learning via Low-rank Reparametrization

Sun Yat-sen University(中山大学);Microsoft Research Asia(微软亚洲研究院)

Da Yu, Huishuai Zhang, Wei Chen, Jian Yin, Tie-Yan Liu

摘要:

我们提出了一个重新参数化方案,以解决在大型神经网络上应用差分私有SGD的挑战,这是1)存储单个梯度的巨大内存成本,2)遭受臭名昭著的维度依赖性的增加的噪声。具体来说,我们重新参数化每个权重矩阵与两个小尺寸的\emph{gradient-carrier}矩阵和一个\emph{residual weight}矩阵。我们认为,这种重新参数化保持向前/向后的过程不变,同时使我们能够计算投影梯度,而无需计算梯度本身。为了学习差分隐私,我们设计了\emph{reparametrized gradient perturbation (RGP)},它扰动梯度载体矩阵上的梯度,并从噪声梯度中重建原始权重的更新。重要的是,我们使用历史更新来找到梯度载体矩阵,其最优性在线性回归下得到严格证明,并通过深度学习任务进行了经验验证。RGP大大降低了内存成本,提高了实用性。例如,我们是第一个能够在BERT模型上应用差分隐私的人,并且使用$ε=8$在四个下游任务上实现了$83.9\%$的平均准确度,与非隐私基线相比,这在$5\%$损失范围内,但隐私泄露风险要低得多。

英文摘要:

We propose a reparametrization scheme to address the challenges of applying differentially private SGD on large neural networks, which are 1) the huge memory cost of storing individual gradients, 2) the added noise suffering notorious dimensional dependence. Specifically, we reparametrize each weight matrix with two \emph{gradient-carrier} matrices of small dimension and a \emph{residual weight} matrix. We argue that such reparametrization keeps the forward/backward process unchanged while enabling us to compute the projected gradient without computing the gradient itself. To learn with differential privacy, we design \emph{reparametrized gradient perturbation (RGP)} that perturbs the gradients on gradient-carrier matrices and reconstructs an update for the original weight from the noisy gradients. Importantly, we use historical updates to find the gradient-carrier matrices, whose optimality is rigorously justified under linear regression and empirically verified with deep learning tasks. RGP significantly reduces the memory cost and improves the utility. For example, we are the first able to apply differential privacy on the BERT model and achieve an average accuracy of $83.9\%$ on four downstream tasks with $ε=8$, which is within $5\%$ loss compared to the non-private baseline but enjoys much lower privacy leakage risk.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑