Scalable Dynamic Matrix Completion for Information Processing and Link Discovery

Recent study by Candes et al. proves that that matrix completion is able to recover the missing entries from a surprising small fraction of random known entries if the matrix admits certain properties. The theoretical advance of the matrix recovery fosters a number of applications such as collaborative filtering, machine learning, and high-dimensional data process. In this project, explore Near-Optimal Matrix Completion (NOMC) for information processing and link discovery in the intelligent communication networks and sensor networks. NOMC advocates matrix completions for Intelligent Link Discovery (ILD) and High-dimensional Object Localization and Tracking (HOLT). Matrix completion for ILD is to predict the subcarrier performance quality from a few observed subcarriers, resulting in high quality link for CR/SDR communications. Due to only sparse subcarrier monitor is required for subcarrier quality prediction, ILD greatly reduces the network communication overhead while other approaches require intensive and continuous subcarrier monitor and data fusion. On the other hand, the matrix completion for HOLT can localize a large number of moving objects and sensors by using a small fraction of sensor location information. This essentially reduces the communication overhead for locating the objects while increasing information values for national security.