That seems like a good approximation of how a real NN would learn a specific task. I'm wondering about a way to have this knowledge, or part of it, carried over

That seems like a good approximation of how a real NN would learn a specific task. I'm wondering about a way to have this knowledge, or part of it, carried over to other tasks, or even slightly different tasks, so that there would be actual learning rather than optimization.