Robust Online Tensor Completion for IoT Streaming Data Recovery.

Area of Science:

• Data Science
• Signal Processing
• Internet of Things (IoT)

Background:

• Reliable data measurement is crucial for Internet of Things (IoT) applications.
• Data missing and corruption are common challenges in practical IoT systems.
• Existing tensor completion methods struggle with the dynamic nature of long data sequences.

Purpose of the Study:

• To develop a robust method for recovering missing IoT measurement data in the presence of outliers.
• To address the limitations of existing tensor completion methods that assume a fixed tensor rank.
• To propose a framework that accounts for time-varying tensor rank and noise in IoT data.

Main Methods:

• Formulating the data recovery problem as a tensor completion (TC) task.
• Developing an updatable framework based on dynamic CANDECOMP/PARAFAC (CP) decomposition.
• Introducing the temporal multi-aspect streaming (T-MUST) algorithm to solve the optimization problem.

Main Results:

• The T-MUST algorithm effectively handles time-varying tensor rank, automatically detecting and tracking rank changes.
• Theoretical analysis confirms T-MUST's geometric convergence rate.
• Empirical validation on synthetic and real-world datasets demonstrates T-MUST's superior efficiency and effectiveness compared to existing methods.

Conclusions:

• T-MUST provides an accurate and robust solution for missing IoT data recovery, outperforming traditional TC methods.
• The proposed framework accommodates the dynamic characteristics of IoT streaming data, including time-varying rank and noise.
• T-MUST offers a significant advancement in ensuring data integrity and reliability for IoT applications.