Can we Generalize and Distribute Private Representation Learning?-猫眼课题宝

DOI：

发表时间：

2020-10

期刊：

影响因子：

通讯作者：

中科院分区：

其他

文献类型：

作者：

Sheikh Shams Azam;Taejin Kim;Seyyedali Hosseinalipour;Carlee Joe-Wong;S. Bagchi;Christopher G. Brinton

我们研究学习私密但信息丰富的表示的问题，即提供有关预期“盟友”目标的信息，同时隐藏敏感的“对手”属性。我们提出了排除-包含生成对抗网络（EIGAN），这是一种广义的私有表示学习（PRL）架构，与现有的 PRL 解决方案不同，它考虑了多个盟友和对手属性。虽然集中聚合的数据集是大多数 PRL 技术的先决条件，但现实世界中的数据通常分散在多个分布式节点中，由于隐私问题而不愿意共享原始数据。我们通过开发 D-EIGAN 来解决这一实际限制，这是第一个分布式 PRL 方法，可以在不传输源数据的情况下学习每个节点的表示。我们从理论上分析了最优 EIGAN 和 D-EIGAN 编码器下对手的行为，以及盟友和对手任务之间的依赖关系对优化目标的影响。我们对各种数据集的实验证明了 EIGAN 在性能、鲁棒性和可扩展性方面的优势。特别是，EIGAN 的性能明显优于之前的最先进技术（提高了 47%），并且 D-EIGAN 在不同网络设置下的性能始终与 EIGAN 相当。我们观察到盟友和对手的性能保持相当稳定（对于 EIGAN），我们增加了 D-EIGAN 下的节点数量。通过这两个实验，我们可以

We study the problem of learning representations that are private yet informative, i.e., provide information about intended “ally” targets while hiding sensitive “adversary” attributes. We propose Exclusion-Inclusion Generative Adversarial Network (EIGAN), a generalized private representation learning (PRL) architecture that accounts for multiple ally and adversary attributes unlike existing PRL solutions. While centrally-aggregated dataset is a prerequisite for most PRL techniques, data in real-world is often siloed across multiple distributed nodes un-willing to share the raw data because of privacy concerns. We address this practical constraint by developing D-EIGAN, the ﬁrst distributed PRL method that learns representations at each node without transmitting the source data. We theoretically analyze the behavior of adversaries under the optimal EIGAN and D-EIGAN encoders and the impact of dependencies among ally and adversary tasks on the optimization objective. Our experiments on various datasets demonstrate the advantages of EIGAN in terms of performance, robustness, and scalability. In partic-ular, EIGAN outperforms the previous state-of-the-art by a signiﬁcant accuracy margin (47% improvement), and D-EIGAN’s performance is consistently on par with EIGAN under diﬀerent network settings. We observe that the performance of the ally and adversary remains reasonably constant (and to EIGAN) we increase the number of nodes under D-EIGAN. the two experiments, we can

Can we Generalize and Distribute Private Representation Learning?

文献摘要

被引文献

相似文献