Sap Se

Philip-William Grassal

Daniel Bernau

Florian Kerschbaum

Martin Haerterich

Jonas Robl

Hannah Keller

Data is received that specifies a bound for an adversarial posterior belief ρthat corresponds to a likelihood to re-identify data points from the dataset based on a differentially private function output. Privacy parameters ε, δ are then calculated based on the received data that govern a differential privacy (DP) algorithm to be applied to a function to be evaluated over a dataset. The calculating is based on a ratio of probabilities distributions of different observations, which are bound by the posterior belief ρas applied to a dataset. The calculated privacy parameters are then used to apply the DP algorithm to the function over the dataset. Related apparatus, systems, techniques and articles are also described.

Interpretability framework for differentially private deep learning

Machine learning model data privacy can be maintained by training a machine learning model forming part of a data science process using data anonymized using each of two or more differential privacy mechanisms. Thereafter, it is determined, for each of the two or more differential privacy mechanisms, a level of accuracy and a level precision when evaluating data with known classifications. Subsequently, using the respective determined levels of precision and accuracy, a mitigation efficiency ratio is determined for each of the two or more differential privacy mechanisms. The differential privacy mechanism having a highest mitigation efficiency ratio is then incorporated into the data science process. Related apparatus, systems, techniques and articles are also described.

Differential privacy to prevent machine learning model membership inference

Methods, systems, and computer-readable storage media for privacy preserving metering is described herein. A resource threshold value associated with anonymizing meter data for resources metered at a first destination is received. Based on a noise scale value and the resource threshold value, an individual inference value of the first destination is computed. The individual inference value defines a probability of distinguishing the first destination as a contributor to a query result based on anonymized meter data of the first destination and other destinations according to the noise scale value. The noise scale value is defined for a processing application. Based on evaluating the individual inference value, it is determined to provide anonymized meter data for metered resources at the first destination. An activation of a communication channel for providing the anonymized meter data for metered resources is triggered. The communication channel is between the first destination and the processing application.

Growing community of inventors

Heidelberg, Germany

Philip-William Grassal