Research
Published research by Caleb Spradlin
cspradlin_identificationofhiggsboson.pdfApplication of Semi-Supervised Machine Learning Methods Towards the Identification of the Higgs Boson
Application of Semi-Supervised Machine Learning Methods Towards the Identification of the Higgs Boson
by Caleb Spradlin
Published in UNC at Asheville Journal for Undergraduate Research May 2020Colliders allow physicists to probe the previously unknown world of sub-atomic physics employing observations of exotic particles through high-energy collisions. Physics communities regularly rely on hand-crafted, high-level features in conjunction with shallow machine-learning packages to accurately identify particles produced in collisions. This process proves excessive and time-consuming. This work provides an innovative means of solving the problem of accurate identification of Higgs boson particles through state-of-the-art, semi-supervised learning methods, and data collected by the European Centre for Nuclear Research. This research demonstrates how using semi-supervised learning techniques, specifically weight-averaged consistencies and data abstraction methods, alleviates the need for fully labeled datasets in accurate identification. Furthermore, it is demonstrated how deep semi-supervised learning models automatically extrapolate high-level features from the data given.