Neuro-Symbolic Approach for Suffix Prediction in Business Process Management
A new research paper has unveiled a method called Neuro-Symbolic Predictive Process Monitoring (PPM), which merges data-driven insights with established knowledge rooted in temporal logic to improve suffix prediction in Business Process Management (BPM). This method takes Linear Temporal Logic over finite traces (LTLf) and incorporates it into the training of autoregressive sequence predictors. It uses a special logical loss function derived from a soft approximation of LTLf semantics, along with the Gumbel-Softmax trick. This ensures that the generated suffixes are accurate and logically sound, addressing the limitations seen in deep learning models that often fail to fulfill essential logical criteria. The method's effectiveness has been confirmed through experiments.
Key facts
- The paper is published on arXiv with ID 2509.00834v2.
- It addresses suffix prediction in Business Process Management (BPM).
- The approach integrates data-driven learning with temporal logic-based prior knowledge.
- It uses Linear Temporal Logic over finite traces (LTLf) in training.
- A differentiable logical loss function is defined using soft approximation of LTLf semantics and the Gumbel-Softmax trick.
- The method combines logical loss with standard predictive losses.
- It aims to generate suffixes that are both accurate and logically consistent.
- Experimental evaluation is included in the paper.
Entities
Institutions
- arXiv