Is there currently a way to write a learning that gets two different error signals as input?
You could pull this commit from the generic learning rule PR, which allowed learning rules to define arbitrary
That doesn’t seem to work properly because the
size_in also defines one dimension of the the
delta signal (which is different).