An Instantiation of the Multiple-Transfer Framework to Reduce Efforts in Context Model Learning for New Users in Smart Homes
Ching Hu Lu, Yi Ting Chiang · 2014
Since a real-life environment may encounter various uncertainties due to its dynamic nature, a smart-home system needs to improve its adaptability in response to the inevitable uncertainties. In this regard, a multi-transfer framework was proposed to keep context models adaptable in order to reduce the efforts in retraining context models in the event of an uncertainty. The framework is used to transfer knowledge from a source domain to a target one by reusing as much information from the source domain. This way, the efforts of training activity models for new users in the target domain can be effectively reduced. This paper presents one instantiation of the framework and its implementation details. The preliminary results show that the effort of training activity models for a new user can be effectively reduced meanwhile maintaining satisfactory performance.