Entropy-regulated cross-modal generative fusion for multimodal network intrusion detection

doi:10.1016/j.inffus.2025.103581

	Entropy-regulated cross-modal generative fusion for multimodal network intrusion detection
	Wang, Xiangbin 1,2; Yuan, Qingjun 1,2; Yu, Wentao 1,3; Meng, Qianwei 1,2; Lu, Siqi 1,2; He, Wenqi 1,2; Gu, Chunxiang 1,2; Wang, Yongjuan 1,2
	2026-02-01
发表期刊	INFORMATION FUSION
ISSN	1566-2535
卷号	126 页码:16
摘要	With the increasing popularity of network encryption protocols, analyzing encrypted traffic has become a significant challenge for network security. In this context, deep learning methods have been widely applied to intrusion detection and traffic classification tasks due to their powerful feature extraction capabilities. However, these methods still face two main limitations: relying on unimodal feature extraction, which ignores the multimodal characteristics of network traffic, or adopting simple static fusion strategies, which may fail to capture the complex semantic associations between different modalities. These limitations make it challenging for models to detect sophisticated attacks concealed within otherwise normal encrypted traffic. To address this challenge, this paper introduces an Entropy-Regulated Cross-Modal Generative Framework for Intrusion Detection (ER-CMGI), which combines the generative power of diffusion models with dynamic information-theoretic optimization techniques. The framework integrates variational autoencoders(VAEs) with a lightweight diffusion model for multimodal feature extraction and generation. It implements an adaptive fusion mechanism using a hybrid entropy-based approach that combines both traditional low-entropy priority and inverse entropy weighting through learnable mixing coefficients. The cross-modal generation consistency is achieved through a lightweight diffusion model, which enables self-supervised learning via direct cross-modal generation and comparison. This design enhances semantic alignment across heterogeneous modalities through cross-modal generative learning. Experimental results show that the proposed model achieves F1 scores of 99.12% and 97.81% on two datasets, respectively. This study presents a dynamically adaptive and entropy-guided framework for intrusion detection in network environments, which shows effectiveness in capturing complex attack patterns. By integrating dynamic feature fusion with cross-modal semantic modeling, the framework enhances detection accuracy and interpretability, offering a promising approach for improving network security under evolving threat scenarios.
关键词	Generative artificial intelligence Intrusion detection system Multimodal fusion Diffusion model Cross-modal representation Differential entropy
DOI	10.1016/j.inffus.2025.103581
收录类别	SCI
语种	英语
资助项目	National Key Research and Development Program of China[2023YFB2705000]
WOS研究方向	Computer Science
WOS类目	Computer Science, Artificial Intelligence ; Computer Science, Theory & Methods
WOS记录号	WOS:001547315300001
出版者	ELSEVIER
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/41981
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Yuan, Qingjun; Wang, Yongjuan
作者单位	1.Informat Engn Univ, Zhengzhou 450001, Henan, Peoples R China 2.Henan Key Lab Network Cryptog Technol, Zhengzhou 450001, Henan, Peoples R China 3.Chinese Acad Sci, Inst Comp Technol, Beijing 100190, Peoples R China
推荐引用方式 GB/T 7714	Wang, Xiangbin,Yuan, Qingjun,Yu, Wentao,et al. Entropy-regulated cross-modal generative fusion for multimodal network intrusion detection[J]. INFORMATION FUSION,2026,126:16.
APA	Wang, Xiangbin.,Yuan, Qingjun.,Yu, Wentao.,Meng, Qianwei.,Lu, Siqi.,...&Wang, Yongjuan.(2026).Entropy-regulated cross-modal generative fusion for multimodal network intrusion detection.INFORMATION FUSION,126,16.
MLA	Wang, Xiangbin,et al."Entropy-regulated cross-modal generative fusion for multimodal network intrusion detection".INFORMATION FUSION 126(2026):16.