Almost invariably, ML procedures provide increased precision compared with hand-crafted human efforts. The question remains as to whether they are also more accurate. For example, the claimed uncertainty band on a parameter of interest may be smaller for the ML procedure, but is it really closer to the true value? Using a historical perspective on how we came to the standard test for statistical rigour in HEP, we extrapolate the test we might need with an ML-based analysis. There will be a focus on times where the statistics were done wrong in an analysis and the consequences, to guide towards new tests for the future. The view will be mostly from a ATLAS/CMS perspective, but will be accessible for any HEP experiment.
Bio: Dr. Brenner (PhD Amsterdam 2017) works on the ATLAS and FASER experiments at CERN. In ATLAS she works on Higgs boson physics as well as the combination and re-interpretations of those measurements for Beyond the Standard Model theories. Dr. Brenner has co-led the first ATLAS EFT interpretation of the combined Higgs data, and created the Langrangian Morphing Method that has been instrumental in these interpretations. After fellowships at DESY and CERN, she is now a senior researcher at Nikhef in Amsterdam. She is the current chair of the ATLAS statistics committee. Dr. Brenner served four years as early career delegate in the European Committee for Future Accelerators, is the chair of Nikhef’s Scientific Advisory Board, and the chair of the PhyStat organisation which coordinates work, meetings, conferences and schools that link physics and statistics communities.