Exponential Separations in Symmetric Neural Networks

Neural Information Processing Systems 

In this work we demonstrate a novel separation between symmetric neural network architectures.