Minimizing Age of Event in Artificial Intelligence of Things

• Published in: ACM transactions on sensor networks Vol. 21; no. 3; pp. 1 - 29
• Main Authors: Huang, Ziyao, Wu, Weiwei, Chau, Vincent, Wu, Kui, Liu, Xiang, Wang, Jianping
• Format: Journal Article
• Language: English
• Published: New York, NY ACM 21.05.2025
• Subjects: Computer systems organization; Data path algorithms; Design and analysis of algorithms; Networks; Real-time systems; Theory of computation; Information freshness; online algorithm; transmission scheduling
• ISSN: 1550-4859; 1550-4867
• DOI: 10.1145/3717834

Information freshness, measured by the Age-of-Information (AoI) metric, is a crucial aspect of conventional network systems. However, the emergence of the Artificial Intelligence of Things (AIoT) introduces unique requirements for assessing information freshness, rendering the traditional AoI definition inadequate. This is because the traditional AoI metric operates under the presumption that each data packet bears equal significance. In contrast, AIoT systems must prioritize the transmission of event summaries from smart IoT devices. To promptly capture events as they occur at the sources, we propose a novel information freshness metric called Age of Event (AoE). Subsequently, we thoroughly investigate the problem of AoE-minimizing transmission scheduling. This issue presents a formidable challenge because the event occurrence pattern can be unpredictable, and more crucially, the base station only becomes aware of these occurrences post-transmission. In response, we formulate algorithms and conduct a theoretical analysis applicable to scenarios characterized by complete, zero, or partial knowledge of event occurrences. Evaluations performed on a real traffic event dataset reveal that even in the absence of complete knowledge, our algorithms exhibit competitive performance when compared against the clairvoyant benchmark and markedly outperform AoI baselines.