Multi Home Transfer Learning for Resident Activity Discovery and Recognition

Activity discovery and recognition can provide unprecedented opportunities for health monitoring, automation, energy efficiency and security. Despite all the potential benefits, in practice we are faced with the main challenge of collecting huge amounts of data for each new physical space in order to carry out the conventional activity discovery algorithms. This results in a prolonged deployment in the real world. More importantly, if we ignore what has been learned in previous spaces, we face redundant computational effort and time investment and we miss the insights gained from past experience that can improve the recognition accuracy. To overcome this problem, we propose a method of transferring the knowledge of learned activities from multiple source physical spaces, e.g. home A and B, to a target physical space, e.g. home C. Our method called Multi Home Transfer Learning (MHTL) is based on a location mining method for target activity discovery, a semi-Em framework for activity mapping, and an ensemble method for label assignment. In this paper we introduce the MHTL methodology. To validate our algorithms, we use the data collected in several smart apartments with different physical layouts.