Preserving Privacy in Smart Manufacturing Analytics

Penn State will partner with Siemens to develop a technique to help manufacturers reduce the risk of privacy breaches during model inversion cyberattacks. In such attacks, an adversary uses model output to steal sensitive raw data.


The Industrial Internet of Things has thrust manufacturers into a data-driven analytics race. Data gathered through advanced sensors can be used to generate models that help decision-makers improve effectiveness and efficiency. But these insights can come at the expense of privacy. Through model inversion cyberattacks, adversaries manipulate the output of these models, reversing the process to extract the sensitive raw data. Current solutions such as anonymizing the data do not provide enough privacy protection.

Proposed Solution

The team will develop a new differential privacy technique — Mosaic Gradient Perturbation (MGP) — that reduces the risk of model inversion attacks


Manufacturing currently lags the health-care field when it comes to privacy protections on artificial intelligence algorithms. Privacy breaches put industrial institutions at a competitive disadvantage. The technique developed through this project will help manufacturers protect privacy while still allowing them to capitalize on the power of data analytics.