Very helpful for a conceptual understanding of neural nets. But why do you label the minimum as “global”? It is only the local minimum for the attractor to which the initial weights belong, unless you are also doing some discrete jumping around in weight-space.