Conference Paper User Adaptation of Convolutional Neural Network for Human Activity Recognition

井上, 中順  ,  Inoue, Nakamasa  ,  篠田, 浩一  ,  Shinoda, Koichi

Abstract—Recently, monitoring human activities using smartphonesensors, such as accelerometers, magnetometers, and gyroscopes,has been proved effective to improve productivity in dailywork. Since human activities differ largely among individuals,it is important to adapt their model to each individual with asmall amount of his/her data. In this paper, we propose a useradaptation method using Learning Hidden Unit Contributions(LHUC) for Convolutional Neural Networks (CNN). It inserts aspecial layer with a small number of free parameters betweeneach of two CNN layers and estimates the free parameters usinga small amount of data. We collected smartphone data of 43hours from 9 users and utilized them to evaluate our method.It improved the recognition performance by 3.0% from a userindependentmodel on average. The largest improvement amongusers was 13.6%.Index Terms—Human activity recognition, User adaptation,Convolutional neural network, Learning hidden unit contributions

Number of accesses :  

Other information