Activity recognition with the aid of unlabeled samples
Donghai Guan, Young-Koo Lee, Sungyoung Lee · 2009
Activity recognition is an important topic in ubiquitous computing. In activity recognition, supervised learning techniques have been widely applied to learn the activity models. However, most of them can only utilize labeled samples for learning even though a large amount of unlabeled samples exist. In our previous work, we have proposed a semi-supervised learning method which can utilize both labeled and unlabeled samples for learning. As an alternative, a new learning method is proposed in this work. It makes use of the unlabeled samples to remove the noises from labeled samples, so that the learning performance is improved. Experimental results show the effectiveness of our method.