ctivity 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 installation 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.