Scholarlink

<p>Next-generation wireless environments face an unprecedented challenge: edge devices generate heterogeneous multimodal data (textual, visual, acoustic, sensory) that traditional cloud-centric frameworks cannot efficiently process due to bandwidth constraints, latency requirements, and privacy concerns. This paper introduces a neural tensor integration for distributed environments (NTIDE) framework that addresses these limitations through three core innovations. First, we develop neural tensor operations where tensor cores are parameterized by deep neural networks rather than fixed decomposition factors, enabling adaptive cross-modal relationship learning. Second, we implement modality-aware weighted aggregation that updates encoder parameters only from nodes possessing each modality while sharing integration module parameters across all nodes, preserving modality-specific characteristics during distributed training. Third, we introduce hierarchical task scheduling that decomposes neural operations into transferable units, enabling dynamic workload redistribution based on real-time node capabilities. Experimental validation on multimodal sentiment analysis demonstrates that NTIDE achieves 0.965 MAE and 75.5% binary classification accuracy while reducing communication overhead by 56.4%, energy consumption by 41.6%, and inference latency by 31.2% compared to baseline approaches. The framework reaches 75% accuracy 20% faster than competing methods, with training time scaling as O(n^0.88) as network size increases. Real-world industrial deployment across 47 heterogeneous edge devices confirmed 93.7% anomaly detection accuracy with 62.3% fewer false alarms than single-modality approaches. These results validate NTIDE's practical effectiveness for distributed multimodal intelligence in resource-constrained wireless networks. Future work will explore quantum computing techniques for tensor operations and integration with emerging 6G wireless standards for ultra-low-latency distributed intelligence.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep neural network-based multi-source heterogeneous data fusion and shared computing model in wireless networks

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Chongke Wang

摘要

Next-generation wireless environments face an unprecedented challenge: edge devices generate heterogeneous multimodal data (textual, visual, acoustic, sensory) that traditional cloud-centric frameworks cannot efficiently process due to bandwidth constraints, latency requirements, and privacy concerns. This paper introduces a neural tensor integration for distributed environments (NTIDE) framework that addresses these limitations through three core innovations. First, we develop neural tensor operations where tensor cores are parameterized by deep neural networks rather than fixed decomposition factors, enabling adaptive cross-modal relationship learning. Second, we implement modality-aware weighted aggregation that updates encoder parameters only from nodes possessing each modality while sharing integration module parameters across all nodes, preserving modality-specific characteristics during distributed training. Third, we introduce hierarchical task scheduling that decomposes neural operations into transferable units, enabling dynamic workload redistribution based on real-time node capabilities. Experimental validation on multimodal sentiment analysis demonstrates that NTIDE achieves 0.965 MAE and 75.5% binary classification accuracy while reducing communication overhead by 56.4%, energy consumption by 41.6%, and inference latency by 31.2% compared to baseline approaches. The framework reaches 75% accuracy 20% faster than competing methods, with training time scaling as O(n^0.88) as network size increases. Real-world industrial deployment across 47 heterogeneous edge devices confirmed 93.7% anomaly detection accuracy with 62.3% fewer false alarms than single-modality approaches. These results validate NTIDE's practical effectiveness for distributed multimodal intelligence in resource-constrained wireless networks. Future work will explore quantum computing techniques for tensor operations and integration with emerging 6G wireless standards for ultra-low-latency distributed intelligence.