Privacy Preserving Solutions for Network Security

Project Description: The idea is to investigate on privacy preserving soliutions for network security, focusing on efficient anomaly/attack detection via privacy preserving machine learning (ML). Recent works have focused on using ML techniques to detect anomalies that might indicate attacks against a network. However, training ML models for anomaly detection requires access to highly sensitive network usage data, which cannot be freely shared among organizations. Even when this hurdle is overcome, organizations who train good anomaly detection models are unwilling to share the model itself. We aim at addressing this issue by employing privacy preserving computation techniques to perform both model training and data classification. The expected outcome of this project is a set of cryptographic protocols that allow for organizations to collaborate in jointly training such models and/or in using them for detecting potential cybersecurity threats without exposing their sensitive internal information.
The collaboration would be carried out between CISAT (and possibly the ML if interested) in ITU, the network security group at University of Brasilia (UnB, Brazil), the cryptography group at University of Washington (UW, USA) and the security group at Monash University (Australia). At UnB, the network security group has expertise in developing techniques for detecting network security threats, with a recent focus on approaches based on machine learning. The research groups at UW and Monash have expertise in efficient privacy preserving machine learning. The group in AU will contribute with expertise in implementing and benchmarking such privacy preserving computation protocols. These four external partners complement very well CISAT's expertise in network security and privacy preserving computation.