Teacher-Assistant Knowledge Distillation Based Indoor Positioning System

Indoor positioning systems have been of great importance, especially for applications that require the precise location objects and users. Convolutional neural network-based indoor (IPS) garnered much interest in recent years due to their ability achieve high accuracy low error, regardless signal fluctuation. Nevertheless, a powerful CNN framework comes with computational cost. Hence, there will be difficulty deploying such system on computationally restricted device. Knowledge distillation has an excellent solution which allows smaller networks imitate performance larger networks. However, problems as degradation student’s performance, occur when far more complex is used train small CNN, because does not fully capture knowledge passed down. In this paper, we implemented teacher-assistant allow simple closely superior scheme. The involves transferring from large pre-trained network by passing through intermediate network. Based our observation, error can reduced up 38.79% implementing framework, while typical only reduce 30.18%.