Unsupervised domain adaptation for sensor data

Speaker: Erica Ye (School of Computer Science, University of St Andrews)

Abstract

Driven by the persuasiveness of sensing technologies in our everyday devices, activity recognition has been widely adopted in many applications, from smart home, personal healthcare, human-computer interaction, to name a few examples. Machine learning, and especially deep learning, has made significant progress in learning sensor features from raw sensing signals with high recognition accuracy. 

However, most techniques need to be trained on a large labelled dataset, which is often difficult to acquire. To tackle this challenge, we have designed several unsupervised domain adaptation techniques to enable transferring the activity knowledge from one environment to many other environments. In this talk, we will introduce our recent developed techniques, including variational Autoencoder-based feature space alignment, and bi-directional generative adversarial networks-based feature space transfer.