Navarin et al. 2017 — Data-Aware LSTM for Remaining Time Prediction

Extends LSTM-PPM to the data-aware remaining-time regression setting — the natural evolution of Polato et al. 2014's classical data-aware model (sources/2014-polato-data-aware-remaining-time) into the neural era.

Contribution

LSTM architecture accepting concatenated activity + event-attribute vectors at each time step.
Handles repeated activities, categorical attributes, and paths with few training examples.
Competitive with or superior to tailored baselines on benchmark logs.

Significance

Together with Tax et al. 2017 (sources/2017-tax-lstm-process-prediction) and Evermann et al. 2017 (sources/2017-evermann-deep-learning-runtime), anchors the 2017 neural-PPM trio — the moment deep learning decisively entered PPM.

Cited by

sources/2023-riess-temporal-loss-remaining-cycle-time — Riess uses the Navarin-style LSTM stack as the baseline architecture when experimenting with temporally weighted L1 losses.
sources/2024-riess-synbps-simulation-framework — cited as canonical PPM remaining-time model motivating the need for data-generating-process control.

Connections

Concepts: concepts/lstm-ppm · concepts/remaining-time-prediction · concepts/trace-encoding