Deceiving Machines: Sabotaging Machine Learning

• Activity Number: 533 - Assuring the Security of Machine Learning and Statistical Methods
• Type: Invited
• Date/Time: Thursday, August 6, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): David Trott*
• Companies: Department of Defense

Abstract

The growing abundance of high-quality data sets, combined with substantial technical developments, has advanced machine learning into a major tool that is employed in a broad array of applications, from cybersecurity to medical diagnosis. Despite the superhuman-like capabilities often ascribed to machine learning, though, it is vulnerable to a variety of manipulations and open to various attacks.