研究人员开发了一种新颖的差分隐私（DP）学习框架，该框架绕过了迭代参数空间优化。该方法不使用私有化梯度，而是采用在公共数据上训练的超网络，从私有数据集的扰动嵌入中生成模型参数。这种方法仅将隐私噪声注入一次到低维表示中，显著减少了其不利影响。理论分析表明，在合成设置中，其效用高于DP-SGD，并且在LoRA微调扩散模型上的实际应用与DP-SGD和其他公共数据引导方法相比，FID得分更低。 AI
影响 这种新的DP学习方法可以实现更有效、更少噪声的私有模型训练，有可能提高敏感应用中的效用。
排序理由 该集群包含一篇详细介绍差分隐私学习新方法的学术论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
Differentially private (DP) training of neural networks is often hindered by the large amount of noise required by gradient-based methods such as DP-SGD, which repeatedly inject high-dimensional noise in parameter space throughout training. In this paper, we propose a new framewo…
arXiv:2606.26772v1 Announce Type: cross Abstract: Differentially private (DP) training of neural networks is often hindered by the large amount of noise required by gradient-based methods such as DP-SGD, which repeatedly inject high-dimensional noise in parameter space throughout…