Abstract
Human activity recognition (HAR) based on sensor data is a significant problem in pervasive computing. In recent years, deep learning has become the dominating approach in this field, due to its high accuracy. However, it is difficult to make accurate identification for the activities of one individual using a model trained on data from other users. The decline on the accuracy of recognition restricts activity recognition in practice. At present, there is little research on the transferring of deep learning model in this field. This is the first time as we known, an empirical study was carried out on deep transfer learning between users with unlabeled data of target. We compared several widely-used algorithms and found that Maximum Mean Discrepancy (MMD) method is most suitable for HAR. We studied the distribution of features generated from sensor data. We improved the existing method from the aspect of features distribution with center loss and get better results. The observations and insights in this study have deepened the understanding of transfer learning in the activity recognition field and provided guidance for further research.
Highlights
Human activity recognition has received considerable attention in the field of pervasive computing
There are several reasons why transfer learning is necessary in Human activity recognition (HAR). (1) When users perform the same activity, the data collected by sensors tends to be quite different because of the differences in physiology and habits (Figure 1). (2) The movement of user will change over time as well
(3) If deep learning wants to be applied for HAR in practice, it will be faced with large number of emerging users
Summary
Human activity recognition has received considerable attention in the field of pervasive computing. What we we want want to to do do in in this this work work is is to to transfer transfer the the deep learning model between users with unlabeled data of target. Deep learning model between users with unlabeled data of target. (1) For the first time in HAR, a large number of empirical studies have been conducted on deep in transfer comparedstudies several have widely-used methods and (1) unsupervised. Inter-class distance (the distance between centers from different classes) is small These two (3) In order toaffect verify improving feature distribution is helpful for unsupervised transferring, distances thethat transferring. Theisinner-class can be decreased by (3) In order to verify that improving feature distribution helpful fordistance unsupervised transferring, Loss, the features from different classes will spread out relatively. The sharpening of the source classification is helpful for transferring
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
