Temporal Convolutional Learning: A New Sequence analysis model to Improve the Performance of Deep Neural Networks in Smart Phone Based Human Activity Recognition

Smart phones have a great potential to be used as low-cost and non-invasive modality for automated recognition of human activities. Similarity of the signals which are captured from different activities may hamper the performance of such systems [1]. Although deep neural networks are effective tools to overcome the above challenge, they do not produce satisfactory results regardless of the temporal dependence of activity signals [2]. The aim of this article is utilizing temporal Convolutional Network (i.e. TCN) as a tool for incorporating temporal nature of activity signals into training process of deep networks. The casual relationship between TCN layers causes that no missing of historical information occurs in this network, therefore it may promote performance of basic deep schemes in recognizing human activities.