Self-supervised learning improves heavy particle mass estimates in physics
Self-Supervised Learning for Robust Resonance Mass Regression in Cascade Decays
Machine Learning
Summary
Finding the mass of heavy particles from their broken pieces is hard because some information can be missing or tricky. The authors use a special kind of AI that first learns on its own to understand patterns, then fine-tunes itself to guess masses better. This method helps them get clearer signals and stay accurate even when data is noisy or changed. Their approach works better than traditional AI trained with labeled data from the start.
What this means in practice
- •For particle physics experiment teams: Improve mass reconstruction accuracy for heavy particles in collider experiments despite missing energy and measurement biases.
- •For machine learning engineers in physics: Use self-supervised pretraining with VICReg to build robust models that generalize better to corrupted or shifted experimental data.
Authors
Ho Fung Tsoi, Alex Yang, Luis Felipe Gutierrez Zagazeta, Shion Chen, Dylan Rankin
Abstract
Reconstructing the mass of a heavy resonance from its decay products with missing energy is one of the central tasks that directly determine the sensitivity in new physics searches at collider experiments. Supervised learning approaches to this problem often struggle to generalize well due to the presence of various systematic uncertainties and distribution shifts. Exhausting all possible variations in the labeled data can be very compute-intensive, while a failure of the model to generalize can corrupt the reconstructed resonance widths that are critical in peak-hunting analyses. In this work, following the foundation model paradigm, we use a self-supervised approach to pre-train a transformer encoder with VICReg to learn an embedding invariant to various corruptions, then fine-tune it for mass regression on a heavy resonance with masses ranging from 2.5 to 6.5 TeV and a SUSY-like cascade decay into an eleven-body final state. We show that the pre-trained model reconstructs sharper resonance peaks and has a more stable performance under various realistic corruptions, compared to a supervised model of the same architecture trained on the same augmented data from scratch.