Using DeepProbLog to perform Complex Event Processing on an Audio Stream

This technical report explores how to combine subsymbolic data, such as audio, with explicit high-level rules to recognise complex events. The neural component extracts information from the signal, while the probabilistic logic layer keeps the rule structure transparent and allows uncertainty to be modelled directly.

The cybersecurity relevance lies in the reasoning pattern. Many modern detection pipelines must merge noisy signals, heterogeneous sources, and semantic constraints. A neuro-symbolic setup like this is therefore interesting for surveillance, event correlation, and alert analysis, especially when the real task is not to classify a single input but to understand a sequence of behaviours.