Preproccessing Sort by the transverse momentum (Creative) normalization Layer Encoder BatchNormalization Layer Learn a graph Run graph updates 4 nodes -> 1 node Decoder 1 node -> 4 nodes Run graph updates Sort again by the transverse momentum OneOff After the Encoder 7 Dense Layers relu activated 50k jets Learning rate of #0.003# Batch size of 100 Train until the loss does not improve for 100 Epochs (but at least for 500 Epochs) Compression size of 9