C2W1: How is each unit responsible for different region?

In the tensorflow implementation of coffee roasting problem, i can see the below plots, which show how each unit is responsible for different bad zones.

What i dont understand is, how does each unit learns to point to different bad zones?
I understand there are different initial weights, but how do the units not overlap ? Can someone please help to clarify?

Hello @Savitha_Suresh,

This is the key to your question. Please check this post out for more explanations, in which you will find a link to another post that shows how you can make them “overlap” (more precisely, how to make the neurons to learn the same thing). If you get the idea of making them “overlap”, then you know how initialization avoid them from “overlapped”.


sure i will check it out, thanks a lot

OK. If you don’t mind to share any new thoughts about the question after you have read them, we can further discuss it.

1 Like