Finding excesses in model parameter space

TL;DR

This paper demonstrates how simulation-based inference can detect weak signals in high-dimensional data without relying on explicit high-level features, improving sensitivity to new physics such as axion-like particles.

Contribution

The work introduces a novel SBI-based approach to identify signals and distinguish them from background events in scenarios with poor detector resolution.

Findings

SBI can detect weak signals without explicit high-level observables.

The method successfully identifies axion-like particles in simulated beam-dump experiments.

It enhances sensitivity to new physics in challenging detection conditions.

Abstract

Simulation-based inference (SBI) makes it possible to infer the parameters of a model from high-dimensional low-level features of the observed events. In this work we show how this method can be used to establish the presence of a weak signal on top of an unknown background, to discard background events and to determine the signal properties. The key idea is to use SBI methods to identify events that are similar to each other in the sense that they agree on the inferred model parameters. We illustrate this method for the case of axion-like particles decaying to photons at beam-dump experiments. For poor detector resolution the diphoton mass cannot be reliably reconstructed, so there is no simple high-level observable that can be used to perform a bump hunt. Since the SBI methods do not require explicit high-level observables, they offer a promising alternative to increase the sensitivity to new physics.