Optimal observables for searches for top-philic resonances in 4t and ttbar production

Given the recent excess of ttbar events, extensions to the Standard Model involving a change in the coupling to the top quark have gained academic interest. While the most likely explanation is a quasi-bound toponium state, the excess could also be due to a new pseudoscalar resonance. In this project, a top-philic pseudoscalar resonance is investigated in four-top events, which provide the phase space to differentiate the two hypotheses. A machine learning classifier is trained to distinguish SM events from BSM events. Furthermore, an uncertainty-aware training is performed to maximize the sensitivity to the resonance.
As a second approach, an MLP is also trained on a ttbar dataset to separate NLO SM events without a bound-state toponium from those with one. The sensitivity to new physics is increased by a differential measurement with the MLP output as an optimal observable.