Whither Reconstructability Analysis
Last modified: June 7, 2006
Reasonable retrospective questions to ask regarding Reconstructability Analysis are:
- has RA contributed anything to scientific areas beyond RA itself?
- how has (and can) RA's development be enhanced by incorporating results from other areas?
- given the more recent development of general methodological areas that
in a fundamental way consider the main problems associated with RA,
why bother with the continued recognition of RA as a distinguishable area of study?
Reconstructability Analysis is "the process of investigating the possibilities of
reconstructing desirable properties of overall systems from knowledge of the
corresponding properties of their various subsystems". Many of the concepts,
problems, and techniques associated with RA had been studied and developed in
less consciously "system" oriented contexts for decades, or even centuries.
But, other than those that had evolved from (or into) mathematical areas,
there was relatively little of this past work that was used outside of the context
in which it was developed.
RA - introduced in 1979 as an identifiable methodological area of study - was
intended to provide a cohesive organizational framework to study problems of
decomposition, simplification, and identification of complex systems,
and to develop results that would be widely applicable.
Two problems in RA that were recognized as fundamental from the beginning were:
1) the characterization of the space of all possible (discrete) models and
development of the ability to generate important subclasses of this space;
2) study and development of appropriate techniques to evaluate models in these subclasses.
While there has been a lot of progress, these problems are
open-ended and not completely resolved. In considering the
significance of RA today, it is relevant that in the intervening years
Computer Science has grown dramatically and much of the methodological
perspective motivating RA has been encompassed in that area.
Not surprisingly, a number of important results associated with RA have
direct applicability in areas that are essentially mainstream Computer Science.
Over the years there have been a reasonable number of technical results,
some very focussed, some of significance mainly to a particular application area,
and some of more general significance and applicability.
There are a number of expository papers and overviews from different
perspectives, where many of the results associated with RA are
elaborated. This paper will refer to some of these but its main focus will
be on the two main problems referred to above and on the benefits that
have resulted from and benefits that can result from more conscious
interaction between RA and other general areas of methodological study such as:
Database Theory, Genetic Algorithms, Neural Nets, Computational Learning Theory,
Graphical Models, Constraint Satisfaction Problems, and Bayesian Networks.