An Active Deep Neural Network Architecture for Human Activity Recognition

Abstract

Wearable technologies such as iPhones and smart watches have become ubiquitous in our modern world. As a result, massive amounts of sensor data are collected on a daily basis. This data can be leveraged to recognize human activity. Namely, neural networks have shown promise in recognizing activity with minimal data transformation and preprocessing. In this work, a novel deep neural network architecture is proposed to do just this. An active learning submodule is also included to reconfigure the algorithm when the setting changes. The effectiveness of this approach is validated using two real world human activity recognition datasets. A brief overview of machine learning, the history of neural networks, and the specific neural network architectures used are also included in this work.