Exploring the Potential of ML Algorithms to Access Sensitive User Data: A New Inference Attack Discovered by Morteza Varasteh
New inference attacks could allow access to sensitive data
Computer scientists are always trying to find ways to prevent the misuse of these algorithms. ML techniques, with their sophisticated data analysis capabilities, can allow third parties to quickly access private data and carry out cyberattacks.
Morteza Varasteh is a researcher from the University of Essex, U.K. He has identified a new type of inference attacks that could compromise confidential data of users and share them with others. This attack is described in a pre-published paper on arXiv. It exploits vertical learning federation (VFL), which is a distributed ML situation in which two parties have different information about the client.
Varasteh explained to Tech Xplore that \”this work is based upon my previous collaboration with a Nokia Bell Labs colleague, where we developed an approach for extracting user private information from a datacenter, referred as the passive part (e.g. an insurance company).\” The passive party works with another data center (e.g. a bank) to develop an ML algorithm.