MadMiner: Machine learning-based inference for particle physics (1907.10621v2)

Abstract: Precision measurements at the LHC often require analyzing high-dimensional event data for subtle kinematic signatures, which is challenging for established analysis methods. Recently, a powerful family of multivariate inference techniques that leverage both matrix element information and machine learning has been developed. This approach neither requires the reduction of high-dimensional data to summary statistics nor any simplifications to the underlying physics or detector response. In this paper we introduce MadMiner, a Python module that streamlines the steps involved in this procedure. Wrapping around MadGraph5_aMC and Pythia 8, it supports almost any physics process and model. To aid phenomenological studies, the tool also wraps around Delphes 3, though it is extendable to a full Geant4-based detector simulation. We demonstrate the use of MadMiner in an example analysis of dimension-six operators in ttH production, finding that the new techniques substantially increase the sensitivity to new physics.

• The paper introduces MadMiner, a framework that integrates machine learning with matrix element information to analyze high-dimensional data in particle physics.
• It employs neural networks to model likelihood functions directly, increasing sensitivity to subtle signals such as those in ttH production studies.
• MadMiner interfaces with simulators like MadGraph5_aMC, Pythia 8, and Delphes 3, streamlining advanced analyses while accurately capturing detector responses.

MadMiner: Machine Learning-Based Inference for Particle Physics

The paper "MadMiner: Machine learning--based inference for particle physics" introduces a tool designed to enhance the precision of measurements and the sensitivity to new physics at the Large Hadron Collider (LHC). This tool, named MadMiner, leverages machine learning in conjunction with matrix element information to directly engage with high-dimensional data, circumventing the need for typical summary statistics or simplified physics assumptions in particle analysis.

Overview

High-energy physics experiments like those conducted at the LHC aim to explore phenomena beyond the Standard Model by examining subtle kinematic signatures. This process often involves constraining a high-dimensional parameter space, characterized by Wilson coefficients in effective field theories (EFTs) or other physics model parameters. Traditional analysis methods frequently reduce these dimensions to summary statistics, but this simplification can result in the loss of valuable information inherent in the complex interaction paths of particle physics processes.

MadMiner, a Python-based framework, is designed to streamline and automate the use of advanced machine learning techniques within particle physics analyses. It interfaces with established simulators such as MadGraph5_aMC and Pythia 8, supporting a wide range of processes and physics models. It can integrate with Delphes 3 for phenomenological detector simulation and is extendable to more detailed simulations like Geant4.

Numerical Results and Claims

In the example analysis of dimension-six operators in $ttH$ production, the novel methods implemented in MadMiner, including techniques like Scandal, Carl, and Alices, showed a significantly increased sensitivity to new physics, thereby broadening the scope of discovery at the LHC. Specifically, the application of these methods allowed for a more efficient use of available data by retaining high-dimensional information, resulting in stronger anticipated discovery potentials compared to traditional methodologies.

Theoretical and Practical Implications

The introduction of MadMiner is particularly impactful in theoretical and practical regards. Theoretically, it provides a more comprehensive framework for addressing the likelihood-free inference challenge, allowing for a direct estimation of complex likelihood surfaces without the loss of information due to dimensional reduction. This approach enables detailed exploration of subtle effects in parameter spaces that would otherwise be inaccessible with coarse-grained methods.

Practically, by utilizing neural networks trained to predict likelihood functions or ratios, MadMiner can offer speed advantages in data processing and can handle the intricacies of detector responses and parton showers without resorting to approximations that might compromise the accuracy.

Future Directions

MadMiner's development fosters potential expansions in its applicability, including interfacing with full Geant4 simulations for experimental collaborations and incorporating broader classes of systematic uncertainties. Furthermore, while the initial focus has been on EFT scenarios, the principles could be extended to various other areas where high-dimensional inference is pivotal.

The continuous iteration on MadMiner's capabilities and its open-source nature is likely to encourage collaborative improvements and adaptations across different particle physics frontiers, progressively refining our understanding of fundamental interactions at high energies. The integration of these advanced inference techniques underscores a trend toward more data-driven, computationally sophisticated methods in the pursuit of new scientific knowledge in particle physics.