Di Francescomarino et al. 2017 — A-priori Knowledge in PPM

Extends LSTM-PPM to incorporate a-priori knowledge about how a case will unfold — e.g., known future constraints like "surgery room unavailable at time T".

Contribution

NOCYCLE — avoids the local-minimum failure mode of LSTM suffix prediction where the model loops indefinitely.
A-PRIORI — takes future constraints as runtime input, steering the LSTM's suffix prediction to remain compatible.
Crucially: a-priori knowledge is handled at inference, not via retraining — so new constraints can be supplied on-the-fly.

Significance

Demonstrates that neural PPM need not be a closed black box; external knowledge (typically LTL-like constraints) can meaningfully improve predictions. Aligns with the APM Manifesto's framed autonomy intuition — predictors constrained by normative frames.

Connections

Concepts: concepts/predictive-process-monitoring · concepts/next-activity-prediction · concepts/lstm-ppm Authors: entities/chiara-di-francescomarino · entities/chiara-ghidini · entities/fabrizio-maggi Related sources: sources/2017-tax-lstm-process-prediction