Neurosymbolic Process Mining

Neurosymbolic learning combines patterns learned from data with explicit knowledge about how a process works. My work brings process structure into the training of neural networks.

DIFF-ERO: A Conformance-Aware Loss for Deep Learning in Process Mining

BPM 2026; proceedings published with a 2027 citation year.

We introduce a training objective that rewards agreement with process structure alongside prediction accuracy. This helps neural networks learn how activities fit together across a process.

Paper ยท Code

All research topics