Interpretable Uncertainty Quantification in AI for HEP

Abstract

Estimating uncertainty is at the core of performing scientific measurements in HEP: a measurement is not useful without an estimate of its uncertainty. The goal of uncertainty quantification (UQ) is inextricably linked to the question, "how do we physically and statistically interpret these uncertainties?" The answer to this question depends not only on the computational task we aim to undertake, but also on the methods we use for that task. For artificial intelligence (AI) applications in HEP, there are several areas where interpretable methods for UQ are essential, including inference, simulation, and control/decision-making. There exist some methods for each of these areas, but they have not yet been demonstrated to be as trustworthy as more traditional approaches currently employed in physics (e.g., non-AI frequentist and Bayesian methods).Shedding light on the questions above requires additional understanding of the interplay of AI systems and uncertainty quantification. We briefly discuss the existing methods in each area and relate them to tasks across HEP. We then discuss recommendations for avenues to pursue to develop the necessary techniques for reliable widespread usage of AI with UQ over the next decade.

Neural Network Based Point Spread Function Deconvolution For Astronomical Applications

“…Like majority of the deep learning frameworks, the DWDN is also a point-prediction method, i.e., the predictions in the network are not associated with error estimates. A large number of UQ approaches are currently applied in the context of AI (artificial intelligence) applications in physics (Chen et al 2022), with associated advantages and disadvantages based on the problem. For instance, a sampling over the network weights (Neal 2012) or an ensemblebased error estimate (Lakshminarayanan et al 2016) may be computationally expensive in our problem, where the individual model training requires 25.5 hours on one Nvidia V100 GPU with 32GB memory.…”