LLM adaptation in PPM

Large language models learn patterns in sequences. We study how to adapt them to sequences of business events, so they can predict what happens next while keeping training costs manageable.

Domain Adaptation of LLMs for Process Data

ML4PM / ICPM 2025 · Process Mining Workshops (2027).

We adapt pretrained language models directly to event sequences and evaluate both individual and combined prediction tasks. The approach updates only a small part of the model, reducing the number of parameters that need to be trained.

Paper · Preprint

Parameter-Efficient and Quantized LLM Fine-Tuning for Predictive Process Monitoring

Journal submission; preprint (2026).

We study how limited model updates and lower-precision computations balance prediction quality, memory use, and training time. The comparison covers predicting the next activity, the next resource, and the remaining time of a case.

Preprint

All research topics