Affilitation: WAPS Main Supervisor: Luc De Raedt Neurosymbolic (NeSy) AI combines neural networks with symbolic logic to produce models that are both data-driven and interpretable (De Raedt et al., 2020). However, traditionally, NeSy approaches require the structures to be given which implies full knowledge about a problem --- a limiting factor to apply NeSy systems in real-world scenarios. A natural solution is to learn structures while simultaneously training the NNs. Building on our recent work (Möller et al., 2025), which demonstrated that training NNs and learning structure simultaneously is possible but was limited to synthetic tasks, the next step is to scale to real-world datasets and larger models. # References De Raedt, L., Dumančić, S., Manhaeve, R., & Marra, G. (2020). _From Statistical Relational to Neuro-Symbolic Artificial Intelligence_ (arXiv:2003.08316). arXiv. Möller, M., Norlander, A., Martires, P. Z. D., & Raedt, L. D. (2025). Neurosymbolic Decision Trees (arXiv:2503.08762). arXiv. https://doi.org/10.48550/arXiv.2503.08762