The University of British Columbia

Published purpose

The Project focuses on building resilient and secure machine learning (ML) techniques for connected health-care systems. With the advent of connected medical devices, there is immense potential for continuous onsite data gathering and analysis. Recent advances in ML techniques have unlocked significant potential for automated detection and diagnosis of medical conditions and disease monitoring, especially for remote and vulnerable populations. Unfortunately, this has also opened the doors to many kinds of malicious security attacks, especially those targeting the ML algorithms via adversarial attacks (i.e., evasion and poisoning attacks). The Project will investigate systematic techniques for making ML algorithms used in the healthcare domain robust and resilient to different kinds of attacks, and to provide early warning of attacks on the medical devices so that suitable actions can be taken. The team will leverage its collective expertise in the domains of cyber-security, dependability, and machine learning, to develop fundamental innovations in this area.