Feasible boundaries for secure Machine Learning models

Machine learning has many limitations and lacks fundamental security standards. Interest is growing across academic researchers as well as industry professionals who all aim to answer the same question: how do we build and deploy machine learning models that are robust, explainable, unbiased, privacy-preserving, and ultimately trustworthy? To address this core issue, a framework was built at Idaho National Laboratories that outlines standards for secure machine learning development. These machine learning pillars provided a basis and guiding methodology for the direction and design of this research, which addresses each of the pillars but focuses on four central data science topics: data types, sourcing, management, and validation.