Comment by andrewjrangel

Can someone help me Grok how the 3D printed Neural Networks back propagate? As I am trying to go through the paper they describe it as a pure optical approach, but what adjusts the refraction elements during the learning process?