Parts | Days | Selection | Search | Updates | Downloads | Help

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 3: Neural Networks I

AKPIK 3.1: Talk

Wednesday, March 22, 2023, 14:00–14:15, ZEU/0118

"Ahead of Time compilation" of Tensorflow models — •Bogdan Wiederspan, Marcel Rieger, and Peter Schleper — University of Hamburg

In a wide range of high-energy particle physics analyses, ML methods have proven as powerful tools to enhance analysis sensitivity. In the past years, various ML applications were also integrated in central CMS workflows, leading to great improvements in reconstruction and object identification efficiencies.

However, the continuation of successful deployments might be limited due to memory and processing time constraints of more advanced models and central infrastructure. A new inference approach for models trained with Tensorflow, based on Ahead-of-time (AOT) compilation is presented that has the potential to drastically reduce memory footprints while preserving and even increasing computational performance.