Scalable AI-assisted Workflow Management for Detector Design Optimization Using Distributed Computing

Researchers have detailed an AI-assisted framework that couples multi-objective Bayesian optimization with the PanDA-iDDS workflow engine to automate detector design optimization across distributed computing resources, according to a paper submitted in 2026 [1]. The framework, presented by Wen Guan and collaborators, targets the high-dimensional parameter spaces that characterize modern detector development [1]. It orchestrates iterative simulations on heterogeneous resources, building on the Production and Distributed Analysis (PanDA) system originally created for the ATLAS experiment at the CERN Large Hadron Collider [1][2]. PanDA’s intelligent Distributed Dispatch and Scheduling component, iDDS, provides the scalable and flexible engine that supports AI and machine-learning-driven workflows [2]. The work was validated using benchmark problems and realistic studies of the ePIC and dRICH detectors planned for the Electron-Ion Collider [1][2]. Results indicate gains in automation, scalability, and efficiency for multi-objective optimization tasks [1]. The initial manuscript was submitted on 31 March 2026 at a size of 1,974 KB; a revised version followed on 25 June 2026 at 2,110 KB [1]. Large-scale distributed computing infrastructures such as the Worldwide LHC Computing Grid require simulation tools that can test new algorithms and optimize resource allocation [6]. A related framework, CGSim, was developed to address limitations in existing simulators, including hardwired algorithms and a lack of real-time monitoring, and has demonstrated near-linear scaling for multi-site simulations with distributed workloads achieving 6x better performance compared to single-site execution [6]. The detector-design framework extends the PanDA ecosystem, which already coordinates workloads across hundreds of sites and thousands of concurrent jobs [6]. By integrating Bayesian optimization, the system iteratively proposes and evaluates design parameters without manual intervention, reducing the time researchers spend on simulation campaign management [1][2]. The approach is positioned as extensible to other computationally intensive scientific applications beyond particle physics [1][2]. The authors note that the combination of a proven workflow engine with AI-driven search addresses a bottleneck in exploring design trade-offs that has grown as detector complexity increases [1].