IEEE Computational Intelligence Magazine - November 2019 - 70

intelligence layer whose definition
and functionality will be extensively
introduced in Section III-C serves as a
framework that binds all the ecosystem's components. Basically, the layer
abstracts the intelligence to a stratum
domain where it can be addressed
and manipulated.
In Figure 4, we provide a description of each of the ecosystem's component characteristics. Later in the article,
some of these concepts will be further
explained and contextualized. Moreover, we will discuss the considerations
of including additional ecosystem
components due to the particular differences between industrial and a consumer systems from a MI perspective.
C. Intelligence Layer

From a software engineering perspective, most applications that currently use
MI integrate intelligence as part of the
application itself. Alternatively, they may

also interface with other applications in
order to acquire the intelligence services
required. This fact implies that MI is part
of the application layer. It works well in
tightly integrated systems and applications with well-defined boundaries that
are not expected to change or be exposed
to high variance. In such systems, the actual need for intelligence remain constant
and the task and goal remains consistent
over time. One example of this type of
application is a chess-playing program.
The rules and context of the game do not
change and the goal is always the same.
For environments where intelligence
needs are constantly changing with lessdefined functions or low coupling to the
application architecture, the limitations
of an integrated intelligence model are
evident; to update the intelligence, the
whole application needs to be updated.
The most direct way to address this
problem is to decouple intelligence from
the application layer and make the intel-

(Consumer) Intelligent Device
Device able to provide a wide set of intelligent
functionalities and that can also take care of
the processed intelligence actuation, by acting
as a control loop and making possible to have
very short reaction times. Through the definition
of the intelligence layer, the device become a
platform where the intelligence can be flexibly
loaded, allowing a shift from a platform with
defined purpose architecture to a machine
intelligence optimized platform.

Intelligent Service
Service's characterization is tailored according to the level of granularity of the
service. Specifically, we outline two different levels of service's granularity.
An Atomic Intelligent Service (AIS) is a service that performs a very specific
operation, the nature of which is not necessarily bound to the execution of an
intelligent task; furthermore, it must not have any external dependency towards
other AISs (e.g., normalization).
A Fine-Grained Intelligent Service (FGIS) is composed by multiple AISs.
From the observer's point of view, the composed FGIS provides intelligent
functionality. This implies that at least one AIS component embeds an intelligent
task (e.g., linear regression).

Intelligence Market
Enables the access to the intelligence from centralized rendezvous locations. It is a
multi-sided platform that provide services of discovery, handshaking, subscription,
delivery of intelligence and safe payments to users, as well as publishing, categorization,
capabilities verification, secure connection and changing. The market may categorize
intelligence according to target application, areas of scope, target platform, and any
other possibility of grouping could be used to facilitate to users the discovery of
the intelligence.
Intelligence Provider
Defines, configures, and initializes the intelligence
services from the intelligence layer. Furthermore, it
handles intelligence algorithms and provides any data
for initializing the local models (pre-computed models,
training data, weights, and bias, etc.). Finally, it directs
the user to other providers with specialized data or
centralized processing capabilities according to the
user requirements.

Intelligent Application
Located in the application layer, stands
as the element that benefits of the output
generated by the intelligent layer, by
receiving as input the outcome of the
intelligent service processing.

Intelligence Orchestration Function
Executes a set of processes aimed at ensuring that the intelligence is
suitably distributed and executed in the underlying system, taking due
account the system architecture requirements. It also ensures an effective
remote execution, update, and configuration of intelligent services,
by making sure that the applications can properly receive and handle
such services. In practical terms, it consists of at least two main
components (Intelligence Service manager and Intelligence Controller),
whose functionality are elucidated in Fig. 6

FIGURE 4 Ecosystem Elements.

ligence as independent of it as possible.
Then, the intelligence becomes a separated stratum that provides services to
applications in the same way that other
layers and platforms provide services (ex.
the Operating System-OS).
An example can be visualized through
an application that can detect a nearby
dangerous object for a child. Traditionally, the application access the device's
camera and GPS to take pictures, and
applies object recognition algorithms
matching the pictures with the location
to detect nearby dangerous objects. If
the application would be built on top of
an intelligence layer, the application
would ask the intelligence layer to provide notifications when dangerous
objects are detected in the vicinity and,
instead of detecting such objects by
itself, it lets the intelligence layer to take
care of such task. The application still
must consider how to display the information and the level of detail required

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2019

IEEE Computational Intelligence Magazine - November 2019

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2019