Semantic Localization for IoT
369
of difference between illuminating a smart light bulb located in your home or one a
thousand miles away. But while the physical location of a web server might affect
the latency of communication or quality of service, it won’t fundamentally change
the content of the hosted page. For an IoT device, its physical relationship with the
world has everything to do with what it can and cannot accomplish.
Any IoT system that seeks to interact with the physical world assumes a model
of space, either explicitly or implicitly. Such a model is a spatial ontology. Broadly,
the subject of ontology from philosophy is a study of the nature of existence, what it
means for something to be and to be something. In computer science, ontology
is usually about association of entities in a model with structured taxonomies,
addressing questions like “is this object an instance or example of that class of
objects?” (taxonomy) or “is this object a part of an instance or example of that class
of objects?” (meronymy) relationships. In prior work, it has been shown that useful
ontologies can be constrained to have a mathematical lattice structure, and that they
thereby acquire enormous algorithmic and formal benefits that can be leveraged to
compose ontologies, perform inference, and check correctness [6–8]. Such ontologies form a subset of commonly used ontology frameworks such as Web Ontology
Language (OWL). Their mathematical structure resembles that of Hindley-Milner
type systems, from which they inherit practical algorithms that scale to very large
numbers of elements. For example, type inference maps into the problem of finding
a fixed point of a monotonic function over a lattice.
Spatial ontologies have more diversity than just choice of coordinate system. A
common dichotomy in ontologies is the distinction between “objects” and “fields”
[9–11]. An “object” is an entity that is distinct, with a clear boundary, and in the
language of [9] is “individual and fully deniable.” Examples of objects include: an
apple, a table, or a flashlight. A “field” describes phenomena without clearly defined
boundaries that are “smooth, continuous and spatially varying” [9]. The magnetic
field emanating from a hand-held bar magnet is a good example of this concept. From
a certain pedantic perspective, the field is present everywhere in the universe, only
its strength is almost everywhere so weak as to be negligible. Some geographical
features like lakes have elements of both objects and fields because it can be hard to
identify where they end.
Spatial ontologies can also vary with respect to their interpretation of entities
with respect to time. SNAP and SPAN are two cooperative ontologies proposed by
Grenon and Smith [10] to capture the distinction between “continuants,” objects
with an identity that persists across time, and “occurants,” processes defined in part
by their beginning and ending. Examples of continuants include the planet earth
or a pair of shoes because it makes sense to consider their spatial properties at
a particular snapshot of time. The same is not true for occurants like a volcanic
eruption or the takeoff of a helicopter. Such occurants unquestionably have a spatial
existence but their reality is best comprehended in four full dimensions; a sequence
of 3D observations misses something essential about the nature of the process. There
is clearly a strong interrelation between SNAP and SPAN ontologies. This point
is not missed by Grenon and Smith, who devote a latter section of their paper to
trans-ontology interrelations between SNAP and SPAN.
369
of difference between illuminating a smart light bulb located in your home or one a
thousand miles away. But while the physical location of a web server might affect
the latency of communication or quality of service, it won’t fundamentally change
the content of the hosted page. For an IoT device, its physical relationship with the
world has everything to do with what it can and cannot accomplish.
Any IoT system that seeks to interact with the physical world assumes a model
of space, either explicitly or implicitly. Such a model is a spatial ontology. Broadly,
the subject of ontology from philosophy is a study of the nature of existence, what it
means for something to be and to be something. In computer science, ontology
is usually about association of entities in a model with structured taxonomies,
addressing questions like “is this object an instance or example of that class of
objects?” (taxonomy) or “is this object a part of an instance or example of that class
of objects?” (meronymy) relationships. In prior work, it has been shown that useful
ontologies can be constrained to have a mathematical lattice structure, and that they
thereby acquire enormous algorithmic and formal benefits that can be leveraged to
compose ontologies, perform inference, and check correctness [6–8]. Such ontologies form a subset of commonly used ontology frameworks such as Web Ontology
Language (OWL). Their mathematical structure resembles that of Hindley-Milner
type systems, from which they inherit practical algorithms that scale to very large
numbers of elements. For example, type inference maps into the problem of finding
a fixed point of a monotonic function over a lattice.
Spatial ontologies have more diversity than just choice of coordinate system. A
common dichotomy in ontologies is the distinction between “objects” and “fields”
[9–11]. An “object” is an entity that is distinct, with a clear boundary, and in the
language of [9] is “individual and fully deniable.” Examples of objects include: an
apple, a table, or a flashlight. A “field” describes phenomena without clearly defined
boundaries that are “smooth, continuous and spatially varying” [9]. The magnetic
field emanating from a hand-held bar magnet is a good example of this concept. From
a certain pedantic perspective, the field is present everywhere in the universe, only
its strength is almost everywhere so weak as to be negligible. Some geographical
features like lakes have elements of both objects and fields because it can be hard to
identify where they end.
Spatial ontologies can also vary with respect to their interpretation of entities
with respect to time. SNAP and SPAN are two cooperative ontologies proposed by
Grenon and Smith [10] to capture the distinction between “continuants,” objects
with an identity that persists across time, and “occurants,” processes defined in part
by their beginning and ending. Examples of continuants include the planet earth
or a pair of shoes because it makes sense to consider their spatial properties at
a particular snapshot of time. The same is not true for occurants like a volcanic
eruption or the takeoff of a helicopter. Such occurants unquestionably have a spatial
existence but their reality is best comprehended in four full dimensions; a sequence
of 3D observations misses something essential about the nature of the process. There
is clearly a strong interrelation between SNAP and SPAN ontologies. This point
is not missed by Grenon and Smith, who devote a latter section of their paper to
trans-ontology interrelations between SNAP and SPAN.