The design and structure of JuliaML has been discussed in many forums and with many participants. Our goal was to unify approaches to modeling and learning, where the difference between models is not as great as one might think. Whether Bayesian or Frequentist, Statistician, Data Scientist, or Machine Learning researcher, we are unified by the underlying math. The design of this ecosystem is an attempt to find the similarities among techniques, and collaborate as much as possible.
See the early issues of Losses.jl (formerly Evizero/LearnBase.jl) for early discussions, as well as issues in OnlineStats.jl, and of course JuliaML/Roadmap.jl (especially issue #8).