Representation Learning for Sensor-based Device Pairing

The emergence of on-body gadgets has introduced a novel research direction: unobtrusive and continuous device pairing. Existing approaches leveraged contextual information collected by sensors to generate secure communication keys. The secret information is represented throught hand-engineered features. In this paper, we propose a learning method based on Siamese neural networks to extract features that signify on-body context while separating off-body devices.