Speaker
Description
Reweighting Monte Carlo (MC) events for alternate benchmarks of beyond standard model (BSM) physics is an effective way to reduce the computational cost of physics searches. However, applicability of reweighting is often constrained by technical limitations. We demonstrate how pre-trained neural networks can be used to obtain fast and reliable reweighting without relying on the full MC machinery. We demonstrate this by implementing a deep neural network for reweighting MC events of singly produced top partners (T)- positively charged hypothetical vector like quarks (VLQs) interacting predominantly with the third generation of standard model quarks. Our implementation allows continuous-valued reweighting of single-T MC events for all three decay modes of the top partner. We also explore the interpretability of our DNN model by exploring the explainability methods of layer-wise relevance propagation and Neural Activity Pattern diagrams and reflect on the networks response to different scenarios of the top partner physics.
| Consider for long presentation | No |
|---|
