Unfolding without Iterations, Adversaries, or Surrogates

Abstract

Correcting measurements for detector effects and constructing appropriate public data representations is a pressing problem in LHC physics. Current methods solve this inverse problem by relying on iterations, minimax optimization, or a surrogate forward mapping. We introduce Adversary-free Unfolding SanS Iteration or Emulation (AUSSIE), which dispenses with these mechanisms while remaining asymptotically correct. AUSSIE replaces the second OmniFold step with a new loss function that directly yields solutions with minimal dependence on the reference simulation. We showcase AUSSIE on various unfolding tasks, including full-phase-space jet substructure.

Figures (13)

• Figure 1: Gaussian toy example: Observable feature $x$ (left) and unfolded feature $z$ (right), comparing AUSSIE and various iterations of OmniFold.
• Figure 2: Regression losses tracked over training: (Top) The $\mathcal{L}_\text{MLC}$ loss, whose optimum enforces the unfolding condition. (Bottom) The $\mathcal{L}_\text{OmniFold}$ loss, which is an inversion of $\mathcal{L}_\text{MLC}$. The black dashed line shows the value of each loss for the optimal solution $\overline{R}\xspace(z)$. Note that $\mathcal{L}_\text{OmniFold}$ is only the direct objective for the first OmniFold iteration.
• Figure 3: Reco-level jet substructure, comparing AUSSIE and various iterations of OmniFold.
• Figure 4: Reco-level classifier score over jet substructure observables, comparing AUSSIE and various iterations of OmniFold.
• Figure 5: Unfolded part-level jet substructure, comparing AUSSIE and various iterations of OmniFold