Visualize: Figure 1 & 2 & 4 & 5 & 8.
Description: "We describe a procedure for explaining neurons in deep representations by identifying compositional logical concepts that closely approximate neuron behavior."