From Logs to Images: A Practice-Informed Exploration of Convolutional Neural Networks for Cybersecurity Analytics

Security Operations Centers (SOCs) generate large volumes of heterogeneous security logs originating from identity platforms, endpoints, networks, and cloud services. While machine learning has been widely applied to cybersecurity analytics, many existing approaches rely on handcrafted features or heavily curated datasets that do not reflect operational SOC realities. At the same time, Convolutional Neural Networks (CNNs) have demonstrated strong performance in domains where data exhibits spatial structure, yet cybersecurity data is inherently non-visual. This paper presents a practice-informed exploration of applying CNNs to cybersecurity analytics by transforming raw security logs into structured, image-like representations. Rather than proposing a novel neural architecture or benchmark-driven evaluation, the focus is placed on representation design, behavioral semantics, and operational relevance. We introduce a conceptual log-toimage transformation pipeline, explain why CNNp inductive biases can capture meaningful security behaviors, and discuss SOC-relevant use cases, limitations, and a roadmap for future evaluation.