The EVENFLOW project announces the publication of its third white paper, titled Neurosymbolic Learning, Forecasting & Verification in EVENFLOW, authored by Nikos Katzouris (NCSRD), Nikos Manginas (NCSRD), Vasilis Manginas (NCSRD), Elina Syrri (NCSRD), Elias Alevizos (NCSRD), Alessio Lomuscio (ICL) and Georgios Paliouras (NCSRD).

This white paper presents key advances achieved within the project in the area of neuro-symbolic AI, focusing on the integration of learning, reasoning, forecasting, and formal verification techniques for trustworthy AI systems.

In this white paper, the work that has been carried out in the project regarding neurosymbolic temporal learning, reasoning, forecasting, and neuro-symbolic verification is summarised. In particular, a brief overview of EVENFLOW techniques is presented for:

• Jointly training neural networks alongside temporal knowledge, in order to align neural perceptive modules with temporal reasoning tasks and to explain the predictions of hybrid neurosymbolic models.
• Forecasting the occurrence of imminent critical events from perceptual data streams, in order to enable proactive decision-making in a robust and transparent fashion.
• Learning interpretable temporal complex event models from sub-symbolic input and utilising such models for detecting and forecasting these events.
• Obtaining formal guarantees on the robustness of hybrid neuro-symbolic systems, that is, formally verifying the property that small perturbations in the sub-symbolic input do not affect the high-level reasoning output.

Read the 3rd white paper here!