-
Differential Privacy for Class-based Data: A Practical Gaussian Mechanism
Authors:
Raksha Ramakrishna,
Anna Scaglione,
Tong Wu,
Nikhil Ravi,
Sean Peisert
Abstract:
In this paper, we present a notion of differential privacy (DP) for data that comes from different classes. Here, the class-membership is private information that needs to be protected. The proposed method is an output perturbation mechanism that adds noise to the release of query response such that the analyst is unable to infer the underlying class-label. The proposed DP method is capable of not…
▽ More
In this paper, we present a notion of differential privacy (DP) for data that comes from different classes. Here, the class-membership is private information that needs to be protected. The proposed method is an output perturbation mechanism that adds noise to the release of query response such that the analyst is unable to infer the underlying class-label. The proposed DP method is capable of not only protecting the privacy of class-based data but also meets quality metrics of accuracy and is computationally efficient and practical. We illustrate the efficacy of the proposed method empirically while outperforming the baseline additive Gaussian noise mechanism. We also examine a real-world application and apply the proposed DP method to the autoregression and moving average (ARMA) forecasting method, protecting the privacy of the underlying data source. Case studies on the real-world advanced metering infrastructure (AMI) measurements of household power consumption validate the excellent performance of the proposed DP method while also satisfying the accuracy of forecasted power consumption measurements.
△ Less
Submitted 8 June, 2023;
originally announced June 2023.
-
Inferring Class Label Distribution of Training Data from Classifiers: An Accuracy-Augmented Meta-Classifier Attack
Authors:
Raksha Ramakrishna,
György Dán
Abstract:
Property inference attacks against machine learning (ML) models aim to infer properties of the training data that are unrelated to the primary task of the model, and have so far been formulated as binary decision problems, i.e., whether or not the training data have a certain property. However, in industrial and healthcare applications, the proportion of labels in the training data is quite often…
▽ More
Property inference attacks against machine learning (ML) models aim to infer properties of the training data that are unrelated to the primary task of the model, and have so far been formulated as binary decision problems, i.e., whether or not the training data have a certain property. However, in industrial and healthcare applications, the proportion of labels in the training data is quite often also considered sensitive information. In this paper we introduce a new type of property inference attack that unlike binary decision problems in literature, aim at inferring the class label distribution of the training data from parameters of ML classifier models. We propose a method based on \emph{shadow training} and a \emph{meta-classifier} trained on the parameters of the shadow classifiers augmented with the accuracy of the classifiers on auxiliary data. We evaluate the proposed approach for ML classifiers with fully connected neural network architectures. We find that the proposed \emph{meta-classifier} attack provides a maximum relative improvement of $52\%$ over state of the art.
△ Less
Submitted 8 November, 2022;
originally announced November 2022.
-
PackageBuilder: From Tuples to Packages
Authors:
Matteo Brucato,
Rahul Ramakrishna,
Azza Abouzied,
Alexandra Meliou
Abstract:
In this demo, we present PackageBuilder, a system that extends database systems to support package queries. A package is a collection of tuples that individually satisfy base constraints and collectively satisfy global constraints. The need for package support arises in a variety of scenarios: For example, in the creation of meal plans, users are not only interested in the nutritional content of i…
▽ More
In this demo, we present PackageBuilder, a system that extends database systems to support package queries. A package is a collection of tuples that individually satisfy base constraints and collectively satisfy global constraints. The need for package support arises in a variety of scenarios: For example, in the creation of meal plans, users are not only interested in the nutritional content of individual meals (base constraints), but also care to specify daily consumption limits and control the balance of the entire plan (global constraints). We introduce PaQL, a declarative SQL-based package query language, and the interface abstractions which allow users to interactively specify package queries and easily navigate through their results. To efficiently evaluate queries, the system employs pruning and heuristics, as well as state-of-the-art constraint optimization solvers. We demonstrate PackageBuilder by allowing attendees to interact with the system's interface, to define PaQL queries and to observe how query evaluation is performed.
△ Less
Submitted 3 July, 2015;
originally announced July 2015.
