Machine Learning Workflows in Production - Deeplearning4j: Open-source, Distributed Deep Learning for the JVM

Machine learning in production happens in five phases. Most machine-learning systems are ad hoc.) Within each phase, we'll explain: Goal Identify a data-intensive business problem and propose a potential machine learning solution. Goal Rapidly explore and de-risk a use case before significant engineering resources are dedicated to it, make "go/no go" recommendation Overlaps with Phase 3 (model training) except here we don't expect a fully tuned model, nor do we expect to produce a reusable software artifact. NOTES: Overlaps with Phase 2 (feasibility study), but here we expect a fully tuned model and a reusable software artifact.