Improving Event Selection with Machine Learning Methods for the $B^\pm \rightarrow K^\pm a,~a\rightarrow \gamma\gamma$ Search at Belle II.

In the search for Axion Like Particles (ALPs) as a extension of the standard model of particle physics, one possibly arising experimental signature is \(B^\pm \rightarrow K^\pm a,~a\rightarrow \gamma\gamma\).
In the context of the search for this signature at the Belle II experiment, this thesis studies an alternative signal selection algorithm on Belle II simulation samples.
Here the signal selection is based on the so-called Punzi-net,  a training of a feed-forward neural network with a loss function inspired by the minimal detectable cross-section. This approach reproduces the signal efficiency from the previous cut-based selection.