Information Technology Reference
In-Depth Information
When the new pattern is constructed it receives the appropriate signals to be able to
reproduce in its turn (an array of constructing signals in the genetic model and a
single self-inspection signal in the self-inspection model).
In the absence of self-inspection signals both models have exactly the same
behavior. Thus, the experiments of the genetic model are characterized by their initial
configurations, in which there won't be any self-inspection signal. Since the latter
cannot be generated from the other ones, these configurations will only be able to
include and/or produce patterns that use the genetic machinery. Conversely, the
experiments of the self-inspection model will include self-inspecting signals in their
initial configurations, and so it will be possible to include and/or produce patterns that
use the self-inspecting machinery.
3.2
Robust Rules
As it has been depicted in previous sections, once defined the basic rules to assure
self-reproduction it is necessary to enhance the rule set in order to achieve robustness
of the patterns. This task supposes an enormous increase in the size of the active
transition table.
Once determined the total number of states
n
and the number of neighbors
k
to be
considered in transitions, the total amount of individual transition rules is determined
by
n
k+1
(in our case, this would suppose more than 68 billion). Nevertheless, most of
these are useless for practical reasons, since the combinations of states that they
represent are not going to be considered. Usually the designer chooses the local
configurations that are relevant to the problem at hand and defines the corresponding
rules, while the other ones are left under a generic “otherwise” condition and produce
a default behavior that is particular of each state. The former are those which we have
called active rules or explicitly defined rules. When designing the basic rules, that is
to say, the ones who assure reproduction in the most favorable conditions provided by
a quiescent environment, the active rules include only need to foresee a quite
predictable and simple interaction between patterns and the surrounding cellular
space. On the contrary, when considering more complex interaction due to the
possibility of encountering scattered unstructured (but not necessarily quiescent) in
the environment of the patterns, the number of situations to be previewed increases
dramatically. This problem has been addressed by Bünzli and Capcarriere [18] using
information-theoretical techniques to design fault-tolerant transition tables in a
systematic way.
Our approach is closer to the logical design of the automaton than to its empirical
behavior. We have used some cellular automata table-editing facilities that help in the
construction of these huge tables. One of these is the possibility of defining and using
variables in the transitions, which allow a small degree of modular abstraction in the
design of the cellular model. A variable is a symbolic object that serves to group
related states in order to program jointly their behavior. For instance, if we define the
CONST variable as the set {4,5,6} to include all three constructing signal states, and
the variable SIGNAL as the set {4,5,6,7,10} to include all transmittable signal states,
the we can define the generic transition (or
varsition
):
1 1 1 SIGNAL 2 2 2 CONST -> SIGNAL
