I wonder if the "not about... the implementation of machine learning" on the help center page is misleading a few people. That phrasing originates in a discussion about acceptable topics, and the term "implementation" still seems problematic.
How about "algorithmic or implementation details, software library or specific hardware-related" kind of phrasing? The idea is to exclude more implementation-related questions, but only the ones that pertain to step-by-step guidance through an implementation or library.
In short, a couple examples under the latter proposal:
- NG: How to implement A* with TensorFlow?
- OK: How to implement Case-Based Reasoning with Neural Networks?
Note the intended use of the "implementation" word in both questions. If there is something here, I still think there is a need for better examples (the OK is too broad yet).
Disclaimer: I am a proponent of having some implementation questions here.