A new multi-task learning based Wi-Fi location approach using L1/2-norm

While many existing multi-task learning based Wi-Fi location approaches pay more attention on the location performance, they generally neglect determining key access points(APs). In order to reduce maintenance cost in complex indoor environment, a new multi-task learning based Wi-Fi location approach is proposed to find the key APs with enough accuracy. First, we introduce extreme learning machine as basic method to establish a new multi-task learning machine. This machine is based on the assumption that the hypotheses learned from a latent feature space, rather than the original high-dimensional feature space, are similar, in which L_1/2-iiorm is utilized to construct L_2-1/2-norm to achieve joint feature selection in multi-task scenario. An alternating optimization method is employed to solve this problem, by iteratively optimizing the latent space and key features. Experiments on real-world indoor localization data are conducted, and the results demonstrate the effectiveness of the proposed approach.