Efficient human activity recognition using hyperdimensional computing

Human activity recognition is a key task of many Internet of Things (IoT) applications to understand underlying contexts and react with the environments. Machine learning is widely exploited to identify the activities from sensor measurements, however, they are often overcomplex to run on less-powerful IoT devices. In this paper, we present an alternative approach to efficiently support the activity recognition tasks using brain-inspired hyperdimensional (HD) computing. We show how the HD computing method can be applied to the recognition problem in IoT systems while improving the accuracy and efficiency. In our evaluation conducted for three practical datasets, the proposed design achieves the speedup of the model training by up to 486x as compared to the state-of-the-art neural network training. In addition, our design improves the performance of the HD-based inference procedure by 7x on a low-power ARM processor.