Energy mover's distance pairs events for LHC model training
Similarity Pairing with Energy Mover's Distance for Self-Supervised Pre-Training at the LHC
Machine Learning
Summary
Training computer models to understand particle collisions at the Large Hadron Collider often needs many fake versions of the same event to learn what stays the same. The authors found a way to match real events that are naturally similar, based on how much energy you'd need to move around to turn one event into another. This means they can train models without making artificial changes to the data. Their method works as well as or better than ones that rely on changing events manually.
What this means in practice
- •For high-energy physics analysis teams: Improve machine learning models for particle collision data without relying on simulated event distortions by using similarity-based event pairing.
- •For machine learning engineers in scientific computing: Train more physics-faithful embeddings in scenarios where handcrafted data augmentations are costly or unreliable by applying energy mover's distance pairing.
Authors
Ho Fung Tsoi, Dylan Rankin
Abstract
Many self-supervised methods for training foundation models at the Large Hadron Collider (LHC) rely on data augmentations to encourage the model to embed events into a representation space invariant to certain physical or detector symmetries. A common challenge arises from the large freedom in choosing a proper set of augmentations on which downstream performance depends. The implementation of augmentations involves either modifying existing events, potentially breaking the event fidelity, or simulating more event variants, which is computationally intensive. In this work, we present a data-driven method of pairing events by their similarity via the energy mover's distance (EMD), which measures how similar two events are in terms of the work required to transform one into the other. With this approach, distinct events are sampled and matched by their similarity to serve as views for learning invariance, keeping the physics content of each event intact without handcrafted distortions. We demonstrate this augmentation-free pairing method by pre-training on QCD jets via self-distillation and show that it can yield semantic jet embeddings with downstream discrimination power comparable to or better than an augmentation-based baseline.