Co-localization from labeled and unlabeled data using graph Laplacian

摘要: This paper addresses the problem of recovering locations both mobile devices and access points from radio signals, a which we call colocalization, by exploiting labeled unlabeled data points. We first propose solution using Latent Semantic Indexing to construct relative when their absolute are unknown. then semi-supervised learning algorithm based on manifold obtain devices. Both solutions finally combined together in terms graph Laplacian. Extensive experiments conducted wireless local-area networks, sensor networks frequency identification networks. The experimental results showthat can achieve high accuracy with much less calibration effort as compared several previous systems.