Privacy-preserving Machine learning (PPML) on trusted hardware



  • Spring 2019-present

Project Description

This project is funded by the Aerospace Corporation which is an American California nonprofit corporation that provides technical guidance on all aspects of space missions to military, civil, and commercial customers. The project tackles problems related to analyzing weather data coming from satellites operated by the National Oceanic and Atmospheric Administration (Noaa). The project tackles research challenges in data management, machine learning, and edge computing.

Currently, weather data are processed by different organizations under different administrative control. For example, some data undergo multiple processing tasks in which it is transferred from the satellite to a ground location in Alaska, and it is further processed in Banglore until it ends up in Maryland. In order to manage the end-to-end analysis of that data, we address the problem of building a dataflow engine in which the processing of the data is tracked as the data moves from one location to another. An interesting research challenge in building that engine is how to do the analysis in a privacy-preserving manner in which the organization processing the data can do the processing without seeing the data owned by another organization. In order to achieve that, we perform the analysis in secure hardware processors (e.g., Intel SGX). Secure hardware provide a guarantee on the privacy of the data/code


  • Tarek Elgamal, Klara Nahrstedt “Serdab: An IoT Framework for Partitioning Neural Networks Computation across Multiple Enclaves” IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID’20), Nov 2-5, 2020, Melbourne, Austrailia.

Funding Agencies

This research is supported and funded by The Aerospace Corporation’s University Partnership Program