I was wondering. What is the purpose of the iron core in a magnetic coil of wire? Then I remembered. What ever it is. It should make sense to a network.
From a little process knowledge in machine learning I speculate that at least in this field you could have two iteration speeds. Maybe the iron core in a magnetic coil wire solves some similar problem. The whole system could perhaps take advantage of two iteration rates?
With this thought I guess you can think of objects in inventions as physical machine learning elements.