Output details

A top-level ontology for smart environments

<02>Machine learning in sensor-driven systems is radically different from traditional problems in that there often exists no high-quality training data or precisely-annotated ground truth from which to learn models. Published in one of the leading pervasive systems journals, this paper consolidates the field by replacing existing ad hoc knowledge- and data-driven approaches with the first formal ontology that hybridises pervasive activities and domain knowledge within a rigorous framework. Reasoning leverages the lattice structure of the ontology to dramatically improve inference in the presence of noisy or limited sensor data, improving core sensor systems tasks such as activity recognition.