Simplifying PELICAN: A Single Hidden Layer Neural Network for Top Tagging

In the world of particle physics, researchers are constantly looking for ways to improve the efficiency and accuracy of their tools. One promising approach is using neural networks, which can help process and analyze large amounts of data generated by particle colliders like the Large Hadron Collider (LHC). However, these networks can be computationally intensive and difficult to implement in real-time systems like the Level-1 trigger at LHC.
To address this challenge, we modified an existing neural network architecture called PELICAN to create a smaller and more interpretable version called nPELICAN𝑁. By omitting certain layers and simplifying the architecture, we were able to reduce the number of parameters while maintaining competitive performance on the top-tagging task.
Our approach involved removing the input embedding layer, all messaging layers, and the output MLP. We then combined the remaining linear equivariant blocks into a single neural network with two layers, separated by a ReLU activation function. The resulting architecture is much simpler than the original PELICAN while still retaining its nonlinearity and ability to perform well on the top-tagging task.
In summary, nPELICAN𝑁 is a tiny neural network that can perform top-tagging tasks in particle physics with competitive accuracy while being computationally efficient and easier to implement than traditional neural networks. This makes it an attractive option for real-time systems like the Level-1 trigger at LHC.