The AI4COPSEC project was presented at the 18th International Conference on Agents and Artificial Intelligence (ICAART 2026), held in Marbella, Spain, from 5 to 7 March 2026.
The presentation was delivered by AI4COPSEC partner Divya Acharya from Simula, and featured the scientific paper “Context-Aware Autoencoders for Anomaly Detection in Maritime Surveillance.” This paper represents an important scientific contribution of the project and a valuable result of the consortium’s research work in the field of AI-based maritime surveillance.
Authored by Divya Acharya, Pierre Bernabé, Antoine Chevrot, Helge Spieker, Arnaud Gotlieb, and Bruno Legeard, the paper addresses the challenge of detecting anomalies in maritime vessel traffic surveillance, which is essential for ensuring safety and security at sea. While autoencoders are widely used for anomaly detection, their ability to identify collective and contextual anomalies remains limited, particularly in the maritime domain, where anomalies often depend on vessel-specific contexts derived from self-reported AIS messages.
To overcome these limitations, the paper proposes a novel context-aware autoencoder approach. By incorporating context-specific thresholds, the method improves anomaly detection accuracy while also reducing computational cost. The study compares four context-aware autoencoder variants with a conventional autoencoder through a case study on fishing status anomalies in maritime surveillance. The results highlight the significant role of context in reconstruction loss and anomaly detection, showing that the context-aware autoencoder outperforms other approaches in detecting anomalies in time series data.
The presentation of this paper at ICAART 2026 marks an important milestone in the project’s dissemination efforts, helping to share AI4COPSEC scientific results with the international research community.
The paper is also available on the Open Access arXiv platform.
More information about the conference presentation is available here.
