End-to-End Optimal Detector Design with Mutual Information Surrogates

TL;DR

This paper tackles end-to-end detector design in high-energy physics by using local deep-learning surrogates to enable differentiable optimization of design parameters $\theta \in \mathbb{R}^{F}$ despite non-differentiable, stochastic forward simulations. It introduces two optimization paradigms: RECO-OPT, which minimizes reconstruction error, and MI-OPT, which maximizes mutual information between true particle features and detector outputs, providing a task-agnostic objective. A five-step optimization loop trains surrogates to approximate the objective and performs gradient-based updates in local neighborhoods, with transfer learning used to propagate knowledge across iterations. Experiments on Geant4-simulated calorimeters show both approaches identify viable designs (increasing scintillator thickness while reducing absorber thickness), with MI-OPT offering close alignment to physics-informed solutions and enhanced generalizability at a potential computational cost. The results demonstrate the practicality and potential of information-theoretic, surrogate-based end-to-end optimization for scientific instrument design, especially when combined with transfer learning to improve data efficiency.

Abstract

We introduce a novel approach for end-to-end black-box optimization of high energy physics (HEP) detectors using local deep learning (DL) surrogates. These surrogates approximate a scalar objective function that encapsulates the complex interplay of particle-matter interactions and physics analysis goals. In addition to a standard reconstruction-based metric commonly used in the field, we investigate the information-theoretic metric of mutual information. Unlike traditional methods, mutual information is inherently task-agnostic, offering a broader optimization paradigm that is less constrained by predefined targets. We demonstrate the effectiveness of our method in a realistic physics analysis scenario: optimizing the thicknesses of calorimeter detector layers based on simulated particle interactions. The surrogate model learns to approximate objective gradients, enabling efficient optimization with respect to energy resolution. Our findings reveal three key insights: (1) end-to-end black-box optimization using local surrogates is a practical and compelling approach for detector design, providing direct optimization of detector parameters in alignment with physics analysis goals; (2) mutual information-based optimization yields design choices that closely match those from state-of-the-art physics-informed methods, indicating that these approaches operate near optimality and reinforcing their reliability in HEP detector design; and (3) information-theoretic methods provide a powerful, generalizable framework for optimizing scientific instruments. By reframing the optimization process through an information-theoretic lens rather than domain-specific heuristics, mutual information enables the exploration of new avenues for discovery beyond conventional approaches.

Figures (13)

• Figure 1: Calorimeter consisting of layers of interleaved lead absorber and lead-tungstate scintillator segments.
• Figure 2: Two developed optimization pipelines: MI-OPT using mutual information as the objective function (left) and REC-OPT using the loss of a reconstruction model for that purpose (right).
• Figure 3: BASE Study: Layer thickness evolution (solid lines) for MI-OPT (left, mutual information maximization in gray dashed) and RECO-OPT (right, reconstruction loss in gray solid, surrogate approximated loss in gray dashed), averaged over three runs for 1-layer (top), 2-layer (middle), and 3-layer (bottom) studies.
• Figure 4: TRANSFER study (700): Re-initialized learning (no transfer) with full sample size and a single layer calorimeter. Comparison of thickness evolutions for MI-OPT (left) and RECO-OPT (right) approaches.
• Figure 5: TRANSFER study (700): Re-initialized learning (no transfer) with full sample size and three layers. Comparison of thickness evolutions for MI-OPT (left) and RECO-OPT (right) approaches.