Increased interest in productionisation created a proliferation of terminology in ML/MLOps. This led to confusion and was recently raised at the MLOps Community slack. The following list is an attempt to clarify these:

ML - creating data defined "declarative" products as opposed to "imperative" software products. Instead of predefining rules, you need to find relationships straight from data.
MLOps - general paradigm around productionised ML (as opposed to academic ML, which only focuses on ML) to do ML in a business context and achieve positive ROI.
MLOps products - OSS or vendor-supplied tools to facilitate parts or whole of the MLOps paradigm.
MLOps architecture - Concrete implementation of infrastructure to facilitate MLOps made up of MLOps products (Buy) and custom implementations (Build).
MLOps processes - Concrete workflows practised on the MLOps architecture to achieve MLOps.
ML orchestration - All of the above: Selecting MLOps products, building MLOps architecture, then running the MLOps processes (but not the actual ML).
ML lifecycle management - High-level MLOps process regarding one model from start to finish.
ML Product Management - Making sure that everyone above is working on the right thing at the right time. Finding out what ML should or shouldn't be done and maintaining the ML lifecycle management.
Our company, Hypergolic, specialises in ML Product Management.
Get in touch for more details!