Pushing the boundaries of knowledge
"A search for new interactions at Belle II using leptons"

This is the official page of the research team "InterLeptons" at the High Energy Physics Institute of the Austrian Academy of Sciences. The team, led by Dr. Gianluca Inguglia, is funded under the grant agreement nr. 947006 of the Starting Grant award offered by the European Research Council (ERC). The research activities of the team will be described and kept up-to-date on these pages.
The aim InterLeptons is to unveil the new physics nature of the so-called flavor anomalies implementing a bottom-up approach based on the studies of data collected at the Belle II experiment, located in the interaction region
of the Super-KEKB collider. The team focuses on final state events containing leptons and a large amount of missing energy. The results of the searches will be interpreted in terms of low mass dark matter, new forces/interactions, and in terms of lepton flavor violating and lepton flavor non-universal couplings.

InterLeptons brings a significant advancement of a new research area in Austria with the potential of revolutionizing particle physics.


InterLeptons News & Co

Back to overview


Punzi-loss and Punzi-net, outperforming standard MVA techniques in the search for new particles of unknown masses

We are happy to announce the submission of our work on the optimization of the search for particles of unknown mass in collider experiments.
The paper, submitted for journal publication, is available on ArXiv and also attached to this page. What we did was to take the inverse of the so-called Punzi figure-of-merit and use that as loss-function in the second step of a neural network originally trained with binary cross-entropy. The result of this is a general improvement of the sensitivity, as minimizing this new loss-function we maximize the Punzi figure-of-merit. The neural network has the capability to learn features of mass hyptheses for which it was not trained and allow us to preform the search of particles of unknown mass optimizing simultaneously our selection for all mass hypothesis. In fact through the use of what we now call a Punzi-net, we are able to find an hyperplane that can separate all signal mass hypotheses and the background simultaneously.
The code is publicly available and can be used by anyone: github.com/feichtip/punzinet
The work was lead by the InterLeptons team members Paul Feichtinger, Huw Haigh and Gianluca Inguglia in collaboration with James Kahn from Helmholtz Artificial Intelligence Center of the Karlsruhe Institute of Technology.


Back to overview

Friends' Pages

Belle II Collaboration
Institute of High Energy Physics (HEPHY)
Academy of Sciences (OEAW)
European Research Council (ERC)
Austrian Science Fund (FWF)
Belle II Facebook
Jennifer 2