Self-supervised Wearable-based Activity Recognition by Learning to Forecast Motion

We propose the use of self-supervised learning for human activity recognition. Our proposed solution consists of two steps. First, the representations of unlabeled input signals are learned by training a deep convolutional neural network to predict the values of accelerometer signals in future time-steps. Then, we freeze the convolution blocks of this network and transfer the weights to our next network aimed at human activity recognition. For this task, we add a number of fully connected layers to the end of the frozen network and train the added layers with labeled accelerometer signals to learn to classify human activities. We evaluate the performance of our method on two publicly available human activity datasets: UCI HAR and MotionSense. The results show that our self-supervised approach outperforms the existing supervised and self-supervised methods to set new state-of-the-art values.