Modeling Abstract Concepts For Internet of Everything: A Cognitive Artificial System
Authors
Abstract
Nowadays, people, things, and processes are all con-nected to internet, and Internet of Everything (IoE) is addressing
them conjointly. It is expected that these entities of IoE will
communicate to one another, and produce a huge amount of
invaluable data, thus generating a need to understand this kind
of data. Computational approaches have shown their capabilities
in extracting valuable information from such data. Usually these
techniques focus on determining concrete concepts (particular
activities such as driving, sitting, or entities like digit ‘1’, ‘2’
etc.). On the contrary, abstract concept identification (actions
such as Motion-state, Static-state, or entity like ‘Number’) is
relatively less explored. In this article we present a methodology
to model abstract representation of concepts without supervision,
along with a mechanism to utilize generated models in an
artificial system. Primarily, we illustrate a Regulated Activation
Networks (RANs) approach that identifies abstract concepts, and
dynamically builds a hierarchy to represent them. Further, we
describe RANs learning, a novel way of associating concepts in
two subsequent layers. We experimentally demonstrate how our
approach is unsupervised, and able to model abstract activity
(static-state and motion-state) of the subjects, using the Smart-
phones Dataset from UCI machine learning repository for Human
Activity Recognition problem.