Neural Circuit Policies - How to Reduce the Size of the Networks

In our self driving project, we have shown that it takes about a hundred thousand neurons and half a million perometers to do end to end learning how to drive from humans. In our neural circuit policies network, we only need 19 notes. And this is an extraordinary dimensionality reduction, because now with 19 nurons, we can get an understanding and the visualization of what the network is doing. Whereas in the huge network model, we have no idea how the net work reaches its conclusions.