EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Adaptive Reinforcement Learning with Cross-Modal Attention for Anomaly Detection

Benjamin Lee, Priya Sharma, Derek Anderson

Computer Life · 2025

DOI: 10.54097/6vbssk40 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer Life, 2025. Verfasst von Benjamin Lee und 2 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Traditional anomaly detection systems often struggle with evolving patterns and multimodal data environments, limiting their effectiveness in dynamic real-world scenarios. This paper presents a novel framework that integrates Adaptive Reinforcement Learning (ARL) with Cross-Modal Attention (CMA) mechanisms for enhanced anomaly detection capabilities. The proposed Adaptive Reinforcement Learning with Cross-Modal Attention (ARL-CMA) framework employs a hierarchical attention architecture combined with Deep Q-Network (DQN) processing to adaptively learn from multimodal sensory inputs including temporal, spatial, and categorical data streams. Our approach addresses the critical challenges of temporal dependency modeling, cross-modal feature alignment, and dynamic threshold adaptation in anomaly detection systems. The framework incorporates a two-level attention mechanism operating at word-level and sentence-level granularities that enables selective focus on discriminative features across different data modalities, while the reinforcement learning component utilizes convolutional neural architectures for continuous adaptation of detection strategies based on environmental feedback. Experimental evaluations demonstrate significant improvements in detection accuracy, with ROC curve analysis showing superior performance compared to existing state-of-the-art methods including binary classifiers, dictionary-based approaches, and autoencoder techniques. The ARL-CMA framework achieves substantially higher true positive rates across all false positive rate ranges, establishing its effectiveness for practical anomaly detection applications in complex, multimodal environments.

Zitieren

APA

Lee, B., Sharma, P., & Anderson, D. (2025). Adaptive Reinforcement Learning with Cross-Modal Attention for Anomaly Detection. Computer Life. https://doi.org/10.54097/6vbssk40

Vancouver

Lee B, Sharma P, Anderson D. Adaptive Reinforcement Learning with Cross-Modal Attention for Anomaly Detection. Computer Life. 2025. doi:10.54097/6vbssk40

BibTeX
@article{Lee2025,
  author = {Lee, Benjamin and Sharma, Priya and Anderson, Derek},
  title = {Adaptive Reinforcement Learning with Cross-Modal Attention for Anomaly Detection},
  journal = {Computer Life},
  year = {2025},
  doi = {10.54097/6vbssk40}
}
Kostenlos starten →