Abstract
The development and deployment of machine learning (ML) systems can beexecuted easily with modern tools, but the process is typically rushed andmeans-to-an-end. The lack of diligence can lead to technical debt, scope creepand misaligned objectives, model misuse and failures, and expensiveconsequences. Engineering systems, on the other hand, follow well-definedprocesses and testing standards to streamline development for high-quality,reliable results. The extreme is spacecraft systems, where mission criticalmeasures and robustness are ingrained in the development process. Drawing onexperience in both spacecraft engineering and ML (from research through productacross domain areas), we have developed a proven systems engineering approachfor machine learning development and deployment. Our "Machine LearningTechnology Readiness Levels" (MLTRL) framework defines a principled process toensure robust, reliable, and responsible systems while being streamlined for MLworkflows, including key distinctions from traditional software engineering.Even more, MLTRL defines a lingua franca for people across teams andorganizations to work collaboratively on artificial intelligence and machinelearning technologies. Here we describe the framework and elucidate it withseveral real world use-cases of developing ML methods from basic researchthrough productization and deployment, in areas such as medical diagnostics,consumer computer vision, satellite imagery, and particle physics.