可用于从非标准环境传感数据中学习准确动态,该模型采用张量分解框架,结合时空正则化和自适应指标权重机制,能有效处理缺失值并保留时空相关性,提升非标准无线传感网数据的表示学习效果。【此简介由AI生成】
Access rate limit exceeded. Please log in to continue.
A Novel Spatiotemporally-Regularized Latent Factorization of Tensors Model for Learning Accurate Dynamics from Nonstandard Environmental Sensing Data
Wireless Sensor Networks (WSNs) are an essential component of intelligent environmental monitoring systems. However, the sensing data collected by WSNs are often nonstandard due to issues such as missing values, sudden sensor failures, or wireless signal dropouts.
To tackle this challenge, we propose SR-LFT (Spatiotemporally-Regularized Latent Factorization of Tensors), a novel representation learning model designed for nonstandard environmental sensing data.
🚀 Key Ideas
-
Latent Factorization of Tensors (LFT) Framework
- Jointly models multiple sensing indicators.
- Introduces a learnable indicator-loss weight to adaptively balance the contribution of each indicator during training.
-
Spatiotemporal Regularization
- Explicitly incorporates both spatial structures and temporal patterns as regularization constraints.
- Enhances the model’s ability to leverage dependencies across multiple sensing indicators.
✨ Contributions
- A new tensor-based latent factorization model with spatiotemporal regularization.
- Adaptive weighting mechanism for heterogeneous sensing indicators.
- Significant improvements in representation learning for nonstandard WSN data.
📊 Experimental Results
We evaluate SR-LFT on two real-world WSN datasets with high missing rates.
- SR-LFT consistently and significantly outperforms six state-of-the-art baselines.
- Demonstrates strong robustness in handling missing values while preserving spatiotemporal correlations.
⚙️ Experimental environment
- Java Development Kit (JDK): 1.8(Need)
Introduction
可用于从非标准环境传感数据中学习准确动态,该模型采用张量分解框架,结合时空正则化和自适应指标权重机制,能有效处理缺失值并保留时空相关性,提升非标准无线传感网数据的表示学习效果。【此简介由AI生成】
Customize your domain