Speaker
Description
The High-Luminosity LHC (HL-LHC) will push the ATLAS experiment into an unprecedented regime of pile-up, data volume, and analysis complexity. In this environment, traditional calibration paradigms, while robust, face increasing pressure in terms of scalability and development effort.
For small-radius jets, the current Monte Carlo (MC) Jet Energy Scale (JES) calibration proceeds through sequential stages: pile-up corrections, MC JES calibration, and Global Sequential Corrections (GSC), including its deep-learning extension (GNNC). Although this structured approach has delivered excellent performance, it relies on multiple independently developed components, each requiring significant time and expert manpower.
We investigate a next-generation calibration strategy based on machine learning that unifies these stages into a single, end-to-end framework. By embedding pile-up mitigation, response correction, and global jet property information into one coherent model, this approach rethinks calibration as a holistic learning problem rather than a sequence of isolated corrections. Such a paradigm shift has the potential to enhance performance, improve robustness under extreme HL-LHC conditions, and substantially reduce development overhead.
This work represents a step toward a more integrated, scalable, and future-ready JES calibration strategy for ATLAS.