The EVENFLOW project is pleased to announce the publication of a new white paper titled On Scalable Neurosymbolic Streaming Analytics, authored by Nikos Giatrakos, ARC partner, and developed within the framework of EVENFLOW. The white paper provides an in-depth overview of a next-generation architecture for real-time artificial intelligence that seamlessly integrates synopsis-driven data management, adaptive neural model training, and neurosymbolic complex event recognition.

Key highlights include:

• SDEaaS (Synopses Data Engine-as-a-Service): A scalable engine that maintains sketches, samples, quantile digests, heavy hitters, and other synopses across thousands of streams—enabling low-latency, resource-efficient analytics.
• SuBiTO (Synopsis-Based Training Optimisation): A control plane that continuously tunes model architectures, training schedules, and synopsis parameters, adapting to concept drift and balancing accuracy vs. latency.
• NeuroFlinkCEP: A neurosymbolic event-processing system that fuses neural detectors with symbolic patterns to recognise complex, multi-step situations in real time.

The white paper also presents nine application domains—from robotics and Industrial IoT to flood response, maritime operations, telecom monitoring, and cybersecurity—illustrating how the NeSy stack delivers scalable, interpretable intelligence in demanding environments.

Τhe publication concludes with a roadmap detailing upcoming advances such as probabilistic CEP, federated synopsis merging, privacy-preserving synopses, and meta-learning for automated synopsis selection.

Read the white paper here!