Security Patterns for Machine Learning: The Data-Oriented Stages
Xinrui Zhang, Jason Jaskolka · 2022
Security in machine learning (ML) is one of the top priorities in many ML-based systems in the field of healthcare, finance, energy, transportation, and cybersecurity. Since developing ML applications requires multidisciplinary effort, it is important to eliminate knowledge mismatch about security among the team members in the early design phase. In this paper, we present a collection of security patterns for the data-oriented stages in the ML workflow, including data collection, data storage, and data preparation. This provides a concise guidance on how to protect each stage from known threats, as well as a communication vocabulary for different roles to consider security without being security experts.