Transfer Learning in Smart Home Scenario
Sonia Sonia, Rashmi Dutta Baruah · 2020
With the development in sensor technology ambiance of human beings are becoming intelligent to cater to the needs and enhance their living standards. As human is dynamic in nature; therefore, a solution should be tailored to the needs of an individual. This requires the capability to understand, analyze and learn the behavior of a human being. To learn human behavior, machine learning algorithms require a sufficient amount of training data. Collection of data and labeling data consumes an ample amount of time. Also, it is not possible to collect data in every possible scenario. To deal with the mentioned problem, in the paper the concept of Transfer learning has been leveraged. The foremost requirement is to calculate the similarity and differences between a selected source domain and a target domain. For the calculation of similarities and differences, multiple parameters are defined in this paper. Multiple experiments in different scenarios were carried out to support the proposed approach. Results obtained show the effects of transfer learning in the domain of smart homes.