RT journal article
T1 PatternStudio: A Neuro-Symbolic Framework for Dynamic and High-Throughput Complex Event Processing
A1 Rosa Bilbao, Jesús
A2 Ingeniería Informática
K1 complex event processing
K1 neuro-symbolic AI
K1 stream processing
K1 Apache Flink
K1 dynamic orchestration
AB Complex Event Processing (CEP) is essential for real-time analytics in domains such as industrial IoT, cybersecurity, and financial monitoring, yet CEP adoption is still hindered by the difficulty of authoring temporal rules and by rigid redeployment workflows. This paper presents PatternStudio, a neuro-symbolic CEP framework that translates natural language specifications into validated event-processing patterns and executes them on a deterministic Apache Flink-based runtime without interrupting service. The generative layer is constrained to produce a typed intermediate representation, while the symboliclayer enforces validation and runtime execution guarantees. We evaluate the prototype as a single-node system-characterization study on commodity hardware representative of edge and near-edge gateways rather than microcontroller-class devices. Under this setting, PatternStudio reaches 47,910 events per second at 250 active rules while maintaining a bounded memory footprint between 1.6 GB and 1.9 GB during the reported runs. Beyond 500 active rules, throughput degradation is driven primarily by CPU saturation and alert amplification, which also explains the sharp increase in tail latency. Additional measurements with parallelism 4, a static baseline, and a two-stage NL-to-IR evaluation further show that the architecture remains functional under partitioned execution, incurs moderate dynamic-orchestration overhead, preserves rule structure reliably under natural-language authoring, and supports interchangeable LLM backends at the semantic front end.
PB MDPI
SN 2624-831X
YR 2026
FD 2026-04-22
LK http://hdl.handle.net/10498/39400
UL http://hdl.handle.net/10498/39400
LA eng
DS Repositorio Institucional de la Universidad de Cádiz
RD 07-nov-2026