Dr. Nikolai Hartmann

Separating signal and background in experimental particle physics with ML

The absence of new physics signals at present collider experiments requires us to increasingly search for small deviations from the standard model. Finding tiny signals, however, requires strong background suppression, which is often only achievable by using as much information as possible - frequently through machine learning methods. In recent years deep learning techniques have become popular. They make use of more low-level information and can sometimes outperform the discrimination power of manually crafted observables. I will introduce both traditional and modern techniques and show examples in R&D benchmarks as well as practical applications at ATLAS and Belle II. Especially for the practical examples, I also want to highlight the approaches taken for estimating background and the treatment of systematic uncertainties.