Second, the properties of our new energy function and the connection to the self-attention mechanism of transformer networks is shown. Finally, we introduce and explain a new PyTorch layer (Hopfield layer), which is built on the insights of our work. ml-jku.github.io/hopfield-lay...