Developed an auto-encoder model equivariant to Lorentz transformations of the input. We find it outperforms graph and convolutional neural networks on jet reconstruction and anomaly detection tasks.
A systematic investigation of evaluation metrics for fast simulations, including two new ones we propose, the Fréchet and kernel physics distances, which we find to be the most sensitive. We also introduce the generative adversarial particle transformer (GAPT) model, which is significantly faster than MPGAN.
Search for events with two high momentum Higgs bosons, using graph neural networks to find Higgs jets. We set the strongest constraints to date on di-Higgs production and the two-vector-boson coupling.
Introduces the message-passing generative adversarial (MPGAN) model and JetNet dataset. We found the physics-informed MPGAN model outperformed all existing point-cloud GANs in simulating high momentum jets.