{"slug": "probabilistic-forecasting-of-business-process-executions-with-neural-temporal", "title": "Probabilistic Forecasting of Business Process Executions with Neural Temporal Point Processes", "summary": "A new arXiv paper (2609.21382v1) presents a generative sequence model for business process forecasting that uses marked temporal point processes to produce full predictive distributions over the next activity and its inter-event time, rather than the single deterministic estimates emitted by mainstream discriminative deep-learning models. The model combines a transformer encoder with a mixture decoder over inter-event times, is trained by exact log-likelihood, and explicitly handles tied timestamps that violate the simple-point-process assumption in real event logs. Across ten public logs, the model matched discriminative baselines on point accuracy, dominated them on calibration and sharpness of remaining-time distributions, and was the cheapest at inference because a full predictive distribution is obtained in a single forward pass without sampling.", "body_md": "arXiv:2609.21382v1 Announce Type: new \nAbstract: Operators of service-based systems act on forecasts of how a running execution will continue, and such a forecast is actionable only if its reliability is known. Mainstream deep-learning models for this task are discriminative and deterministic: they emit a single next activity and a single remaining-time estimate, without a distribution to reason over. We instead cast the problem as generative sequence modelling with marked temporal point processes, which define a joint density over the next mark and its inter-event time and therefore deliver predictive distributions by construction. Real event logs violate the simple-point-process assumption these models rest on, since consecutive events frequently carry identical timestamps; we handle such ties explicitly and combine a transformer encoder with a mixture decoder over inter-event times, trained by exact log-likelihood. On ten public logs, the resulting model matches discriminative baselines on point accuracy, dominates them on the calibration and sharpness of remaining-time distributions, and is the cheapest at inference, since a full predictive distribution is obtained in a single forward pass without sampling.", "url": "https://wpnews.pro/news/probabilistic-forecasting-of-business-process-executions-with-neural-temporal", "canonical_source": "https://www.machinebrief.com/news/probabilistic-forecasting-of-business-process-executions-wit-wv4v", "published_at": "2026-09-21 04:00:00+00:00", "updated_at": "2026-09-21 04:25:18.403491+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "neural-networks"], "entities": ["arXiv", "transformer"], "alternates": {"html": "https://wpnews.pro/news/probabilistic-forecasting-of-business-process-executions-with-neural-temporal", "markdown": "https://wpnews.pro/news/probabilistic-forecasting-of-business-process-executions-with-neural-temporal.md", "text": "https://wpnews.pro/news/probabilistic-forecasting-of-business-process-executions-with-neural-temporal.txt", "jsonld": "https://wpnews.pro/news/probabilistic-forecasting-of-business-process-executions-with-neural-temporal.jsonld"}}