Named for Leonhard Euler, or Leonard Hofstader. You choose.
Leonard is a neural network built to learn how to take the coordinate of a point in space and construct a set of rotation angles that map to a kinematic chain to produce an appropriate end effector position.
The general idea is that
$f^{−1}(f(x))=x$ , so
loss = lossfn($f^{-1}(f(x))$ ,$x$ )
Where$f(x)$
is the function that the neural network is trying to learn.
I am calling this Inverse Functional Training, or IFT. I am sure it has a name already.