Systems, Man & Cybernetics - April 2015 - 22

have a good ontology for solving many difficult problems.
learning is actually a governance process in which the
In fact, self-organization or unsupervised learning ability
learner is the object and the supervisor is the subject.
is another property of a governable or a self-governable
During learning, the subject usually provides a teacher
distributed system.
signal, which can be a desired output or a reward/penLet us examine the human brain. Although we still do
alty to the object for any given input, so that the object
not understand the self-organization mechanism inside
knows what is correct and what is wrong. Through
the brain very well, it seems that the brain is a swarm of
learning, the states (parameters) of all components
neurons, and during the self-organization process, each
inside the object as well as the connections between
neuron learns the behaviors of the leader(s) in its neighborthe components should be adjusted iteratively so that
hood and updates its state and parameters accordingly.
the components can work together to realize different
As a whole, the brain can become well-organized without
concepts and all the concepts can work together to realinstructions from an external governor.
ize the ontology. After learning, it is expected that the
The basic rule for self-organization is the leaderobject can provide correct outputs for the same or simifollowing principle, given as follows. For any unit n 0
lar inputs in a given domain.
We can consider people in a society as the learners and
in a distributed system, if there are some other more
the governor as the supervisor. The governor may repeatsuccessful units n 1, n 2, f, n r in the neighborhood of n 0
edly broadcast regulations (e.g., ethics, laws, and good
under some given conditions c 1, c 2, f, c q, n 0 will try to
behavior patterns) to the people via different media (e.g.,
update itself as follows:
books, newspapers, television, radio, schools, and churchp n (t + 1) = f [ p n (t), p n (t), f, p n (t), c 1, f, c q],
(1)
es), and each person will understand what is correct
and what is wrong for any given
where p ni (t) is the parameter
situation. Eventually, an ontology
vector (or set) at time t for repwill be constructed in the brain of
resenting n i, and f is an updateach person, and the person will
ing function that may (with a
Most problems
become well behaved. Because
high probability) make n 0 more
of the learning ability of human
successful under the same (or
studied in
brains, people can be easily govsimilar) conditions.
conventional control
erned via education.
The well-known self-organizing
A society of well-educated peofeature
map algorithm is a special
theory are related to
ple is also easy to govern because
case [23], in which the winner for
Type_000 cybernetics.
people usually think and behave in
a given input datum (condition) is
a similar way under similar condiconsidered the most successful
tions [20]. Thus, a smart way to
unit, and all units around it just folgovern a society well is to provide
low it. Particle swarm optimization
people with good education opportunities and to reward
(PSO) is another special case [24], in which f is realized
well-behaved people.
in two steps: 1) to update the velocity and 2) to update the
The learning mechanism inside a human brain or a disposition of the unit. In most existing PSO algorithms, the
tributed system in general is not yet completely known,
functions used for updating the velocity and the position
although many algorithms have been studied in the context
are linear, and only one leader is considered in a certain
of neural computing [21]. Among them, the well-known
neighborhood.
back-propagation (BP) algorithm may actually be used in the
It is possible that each neuron in a brain updates
brain [22]. However, algorithms such as BP are not enough
itself based on the leader-following principle and tries to
because they do not reorganize the neurons to construct
become an expert for representing certain sensory patvarious concepts nor connect the concepts to construct the
terns. Each neuron group may also update based on the
ontology. To construct a good ontology for a distributed sysleader-following principle and try to become an expert
tem and to make it governable or even self-governable, more
representing a certain concept. As a whole, the brain can
efficient and effective algorithms should be developed.
become well organized, and an ontology can be constructed through self-organization.
Similarly, human societies also self-organize. In a sociUnsupervised Learning
ety,
people usually try to improve themselves by imitating
For the distributed system to become governable or selfpeople who are more successful. The imitation process
governable, supervised learning is important, but the
also follows the aforementioned leader-following pringovernor may not exist or may not be able to provide the
ciple. It is worthwhile to note that leader following is not
teacher signals for all situations. For instance, in the case
just copying but improving based on additional informawhere the system is a country like the United States, even
tion carried by the leaders. By imitating other people (or
if an external governor does not exist, people in the coungroups), people (or groups) try to become more successful
try can be well organized, and, as a whole, the country can
0

IEEE Systems, Man, & Cybernetics Magazine A pr il 2015

Table of Contents for the Digital Edition of Systems, Man & Cybernetics - April 2015