Robust Learning and Reasoning for Complex Event Forecasting (“EVENFLOW”) (started September 2022, duration 3 years)

The demand for effective solutions in temporal, information-intensive tasks is rising in the rapidly advancing landscape of AI-driven applications. To address this need, EVENFLOW envisions the development of Neuro-Symbolic (Ne-Sy) learning techniques for complex event forecasting. These techniques amalgamate the strengths of deep learning and logic-based learning, embedding them into neuro-symbolic forecasting models.

The approach involves: (i) Ne-Sy techniques to construct event-based features from perception-level data streams. (ii) Integrate symbolic learning and reasoning, generating interpretable patterns for proactive forecasting in critical situations.

The resulting models from EVENFLOW are designed to be interpretable, scalable, and capable of handling dynamic data. SAIL collaborators will work extensively on the verification of Ne-Sy techniques used in EVENFLOW.

A myriad of AI applications relies on the capability to forecast critical events accurately in evolving information scenarios. EVENFLOW, through its Ne-Sy learning techniques, merges deep learning and logic-based reasoning to create neuro-symbolic forecasting models.

Personalised Medicine

EVENFLOW has an emphasis on Personalized Medicine, delving into the intricate patterns of tumour growth in individuals. Utilizing advanced deep generative models, such as Variational Autoencoders (VAEs), EVENFLOW addresses challenges from incomplete and sporadic patient data. By continuously monitoring tumour changes, EVENFLOW enables the prediction of a patient’s condition during cancer progression. This forecasting empowers doctors to tailor treatments, significantly improving patient outcomes.

Industry 4.0

In the realm of factory automation, EVENFLOW equips autonomous robots with forecasting capabilities. Focusing on smart factories, EVENFLOW seeks to enhance the safety and efficiency of autonomous mobile robots (AMRs). By developing predictive models to analyse high-frequency sensor data, EVENFLOW aims to pre-emptively forecast issues, ensuring smoother and safer factory operations.

Infrastructure Life Cycle Assessment

EVENFLOW extends its application to Infrastructure Life Cycle Assessment, envisioning real-time monitoring of underground water pipe networks. EVENFLOW aims to use advanced complex event forecasting techniques to predict and prevent malfunctions in these critical infrastructures. The integration of neuro-symbolic learning facilitates the development of a digital twin for pipe networks, enabling accurate forecasting of lifecycle assessment states and incidents.

In all three use cases, EVENFLOW’s technology showcases its prowess in forecasting critical events, providing actionable insights for personalized medicine, autonomous robotics, and infrastructure management.


In EVENFLOW, our pivotal role involves the development of cutting-edge verification methods tailored for neural-symbolic systems handling complex events. We aim to advance Ne-Sy verification, developing robust verification techniques that enhance the reliability and interpretability of forecasting models. The objective is to implement carefully selected verification methods, integrating them into tools that support complex event systems. A crucial aspect of our contribution involves implementing functionality dedicated to counter-example generation within the complex event pipeline. By focusing on these key areas, we fortify EVENFLOW’s capabilities, ensuring the accuracy and robustness of Ne-Sy forecasting models in diverse and dynamic applications.

This project is funded by the EU, under Horizon Europe grant agreement No 101070430