Random forests are a popular method for classification and regression due to their versatility. However, this flexibility can come at the cost of user privacy, since training random forests requires multiple data queries, often on small, identifiable subsets of the training data. Privatizing these queries typically comes at a high utility cost, in large part because we are privatizing queries on small subsets of the data, which are easily corrupted by added noise. In this paper, we propose DiPriMe forests, a novel tree-based ensemble method for differentially private regression and classification, which is appropriate for real or categorical covariates. We generate splits using a differentially private version of the median, which encourages balanced leaf nodes. By avoiding low occupancy leaf nodes, we avoid high signal-to-noise ratios when privatizing the leaf node sufficient statistics. We show theoretically and empirically that the resulting algorithm exhibits high utility, while ensuring differential privacy.
Non-parametric Differentially Private Confidence Intervals for the Median
Institute for Employment Research(就业研究所);University of Maryland(马里兰大学);University of Pennsylvania(宾夕法尼亚大学);Apple(苹果公司);Harvard John A. Paulson School of Engineering and Applied Sciences(哈佛大学约翰·A·保尔森工程与应用科学学院);Boston University(波士顿大学)
School of Mathematics and Statistics(数学与统计学学院);Beijing Jiaotong University(北京交通大学);Department of Mathematical and Statistical Sciences(数学与统计学系);University of Alberta(阿尔伯塔大学)
University of Massachusetts, Amherst(马萨诸塞大学阿默斯特分校);Google Research NYC(谷歌研究院纽约分部)
差分隐私被广泛采用,以提供可证明的隐私保证在数据分析。我们考虑的问题相结合的公共和私人数据(更一般地说,数据与异构的隐私需求)估计汇总统计。我们引入了一个混合估计的均值优化,以最小化方差。我们认为,我们的机制是更可取的技术,保护个人的隐私,通过子采样数据成比例的用户的隐私需求。同样,我们提出了一个混合中位数估计指数机制的基础上。我们将我们的机制与Jorgensen et al. [2015]中提出的方法进行了比较。我们的实验提供了经验证据,表明我们的机制往往优于基线方法。
Differentially Private Algorithms for Clustering with Stability Assumptions
Ben-Gurion University of the Negev(内盖夫本-古里安大学)
本文研究了在输入稳定性假设下的差分私有聚类问题。尽管关于差分隐私的研究越来越多,特别是关于差异隐私聚类的研究,但只有三项研究(Nissim et al. 2007,Wang et al. 2015,Huang et al. 2018)研究了私下聚类“好”k均值实例的问题,这三个都依赖于样本,聚合框架和所有三个测量实用程序在真正的集群中心和返回的中心之间的Wasserstein距离私有算法在这项工作中,我们改进了这一系列的多轴工作。我们提出了一个简单得多的算法聚类稳定的输入(不依赖于样本和聚合框架),并分析其效用的Wasserstein距离和k-means成本。此外,我们的算法有直接的类似物“好”的k-中位数的实例和本地模型的差分隐私。
Passive Inference Attacks on Split Learning via Adversarial Regularization
National University of Singapore(新加坡国立大学);Massachusetts Institute of Technology(麻省理工学院);Mohamed bin Zayed University of Artificial Intelligence(穆罕默德·本·扎耶德人工智能大学);Renmin University of China(中国人民大学)
Metric Embedding Initialization-Based Differentially Private and Explainable Graph Clustering
Graduate School of Arts and Sciences(艺术与科学研究生院);Columbia University, New York, USA(哥伦比亚大学);School of Artificial Intelligence(人工智能学院);Hebei Institute of Communications, Shijiazhuang, PR China(河北通信学院)
HLF-FSL. A Decentralized Federated Split Learning Solution for IoT on Hyperledger Fabric
atlanTTic Research Center, Information & Computing Lab (ICLAB)(atlanTTic研究所以及信息与计算实验室);Telecommunication Engineering School, Universidade de Vigo(电信工程学院,乌维戈大学);CUD Escuela Naval Militar Universidade de Vigo / Escuela Naval Militar(CUD海军学院,乌维戈大学 / 海军学院)
Optimizing Model Splitting and Device Task Assignment for Deceptive Signal Assisted Private Multi-hop Split Learning
Department of Electrical and Computer Engineering, University of Miami(电子与计算机工程系,迈阿密大学);Department of Computer Science, NC State University(计算机科学系,北卡罗来纳州立大学);Department of Electrical and Computer Engineering, Princeton University(电子与计算机工程系,普林斯顿大学);Frost Institute for Data Science and Computing, University of Miami(数据科学与计算弗罗斯特研究所,迈阿密大学)
Noninteractive Locally Private Learning of Linear Models via Polynomial Approximations
State University of New York at Buffalo(纽约州立大学布法罗分校);Boston University(波士顿大学)
最小化凸风险函数是许多基本学习算法的主要步骤。我们研究协议凸优化,可证明泄漏很少的个别数据点,构成的损失函数。具体来说,我们考虑在本地模型中运行的差分私有算法,其中每个数据记录存储在单独的用户设备上,并且由这些设备在本地执行随机化。我们给出了\emph{noninteractive} LDP凸优化的新协议,协议只需要每个用户向不可信聚合器提交一份随机报告。我们研究了我们的算法在预期损失方面的性能-无论是在手头的数据集(经验风险)还是在假设我们的数据集来自的更大的人群中。我们的误差范围取决于个人对预期损失的贡献形式。对于\emph{generalized linear losses}(如铰链和逻辑损失)的情况,我们给出了一个LDP算法,其样本复杂度仅在维度$p$上是线性的,在其他项(隐私参数$ε$和$δ$,以及期望的超额风险$α$)上是拟多项式的.这是第一个算法的非光滑损失的次指数依赖于$p$。对于欧氏中值问题,其中的损失是给定的欧氏距离到一个给定的数据点,我们给出了一个协议的样本复杂性增长准多项式在$p$。这是第一个协议的次指数依赖于$p$的损失,不是一个广义的线性损失。我们的铰链损失的结果是基于一种技术,称为多项式的内积近似,这可能适用于其他问题。我们的广义线性损失和欧氏中位数的结果是基于新的减少铰链损失的情况下。
PrivDFS: Private Inference via Distributed Feature Sharing against Data Reconstruction Attacks
College of Computer Science and Artificial Intelligence, Fudan University(计算机科学与人工智能学院,复旦大学);Department of Artificial Intelligence, Ajou University(人工智能系,庆应大学)
Privacy-Preserving Split Learning with Vision Transformers using Patch-Wise Random and Noisy CutMix
Yonsei University(延世大学);Duke University(杜克大学);Singapore University of Technology and Design(新加坡科技设计大学);Massachusetts Institute of Technology(麻省理工学院);University of Oulu(奥卢大学)
本文提出 DP-CutMixSL 框架,通过在分割学习中注入高斯噪声并跨客户端随机混合粉碎数据块,以差分隐私机制增强 ViT 训练时的隐私保护,同时提升分类准确性。