Weakly Supervised Human Activity Recognition From Wearable Sensors by Recurrent Attention Learning

Traditional methods of human activity recognition from wearable sensors rely on good training datasets in which thousands of training sequences should be carefully labeled. However, unlike images or videos which can be easily classified by human beings, strictly labeling such sequences of sensor data needs much more manpower and computing resources. In this paper, we present a new weakly supervised human activity recognition model based on recurrent attention learning, in which an agent is trained to extract information from weakly labeled sensor data by adaptively selecting a sequence of locations. Since, the model is non-differentiable and multiple activities may occur in a sequence of sensor data, it is trained by reinforcement learning with novel reward strategies. We evaluated our model on the traditional UCI HAR dataset and our collected weakly labeled dataset. The experimental results show that our model is superior to the traditional CNN model and the DeepConvLSTM model on both datasets.