Created: April 26, 1998
Alex Kosorukoff and David Goldberg Evolutionary computation as a form of organization (Best MPP paper at GECCO-2002)
Traditional areas of application of genetic algorithms (GA) are engineering and technology. Success of genetic algorithms there is well known. This paper explores the use of genetic algorithms as models to influence the design of organization. In particular, we outline the concept of evolutionary organization process based on two recent cases: the Teamwork for a Quality Education (TQE) and Free Knowledge Exchange (FKE) projects.
The distinguishing feature of both projects is that computational evolutionary processes influence the organizational environment, providing the structure of interactions of people and facilitating their communication. In both cases, the organizational structure and people become directly involved into the evolutionary process integrating the power of evolutionary computation with the competence of participating human beings.
Alex Kosorukoff Using incremental evaluation and adaptive choice of operators in a genetic algorithm
Genetic algorithm (GA) is an optimization procedure that uses a simplified model of natural evolution to achieve an objective of optimization represented by a fitness function. The number of fitness function evaluations is a major factor determining the efficiency of a GA. This paper addresses the issue of incremental evaluation which, if applicable, substantially reduces the complexity of evaluation. Incremental evaluation can be used together with evaluation relaxation methods using quick estimates of the real value of the fitness function. Using incremental evaluation in a genetic algorithm puts more emphasis on the rational choice of trade-off between the amounts of time spent on local and global search. This paper contains utility analysis of crossover and mutation operators for one-max problem in terms of their costs and innovation rates. It shows the utilities of these innovation operators change significantly during the run. Experiments with adaptive choice strategy maximizing the expected utility of operators show significantly faster convergence of the algorithm to the optimum solution.
Alex Kosorukoff, Jay Mittenthal and David Goldberg Modeling of evolution of signaling networks in living cells by evolutionary computation
This work is an attempt to model the evolution of signaling pathways in living cells with evolutionary computation under different selection criteria and to compare pathways evolved in such a way with those we encounter in nature.
It proposes 5-level evolutionary procedure for this purpose, that can be thought as several co-operating genetic algorithms working in parallel on different levels of an abstraction hierarchy corresponding to the conceptual structure of an evolving object.
On one hand, this work allows to get some insight into the process of emergence of signaling networks within cells.
On the other hand, this is an example of application of genetic algorithm for redesign of a system that was already designed by natural selection. It allows to compare the results produced by computational model of evolution (GA) with the results of natural evolutionary process, which may lead to better understanding of relationship between natural and computational evolution. We can see how close our GAs are to their natural prototype and can get some new understanding of GAs and ideas on their improvement.
Alex Kosorukoff and David E. Goldberg Genetic Algorithms for Social Innovation and Creativity
Since their invention, genetic algorithms have been used primarily for solving problems in different areas of engineering and technology. In most of these areas genetic algorithms were applied successfully and shifted the frontier between what is possible and what is impossible, solving problems that were even hard to approach with conventional deterministic methods.
Recent social applications of genetic algorithms challenge our usual way of thinking about social systems. Instead of old concepts of social engineering, based on deterministic mechanical constructivism, genetic algorithms and other methods of evolutionary computation open new approaches to the same issue. These methods allow us to substitute fixed structures of our organizations with free forms which are soft and evolving. This paper contains two examples, and some conclusions from this case study. The main conclusion is that GAs have already passed beyond only their technical applications to artificial systems, they are ready to come into the real world of living systems and even have made the first steps in this direction.
Human Based Genetic Algorithm: processing natural language strings for effective knowledge management and innovation (Another copy)
Genetic algorithms (GA) are search procedures learned from Nature and based on mechanics of natural selection and genetics.
In this paper a new kind of GA is presented. It processes natural language strings and organizes knowledge flows among human individuals.
It organizes individuals and uses their ability to perform intelligent crossover and selection operators on existing knowledge.
Genetic algorithms that use human judgment to evaluate solutions are known as interactive genetic algorithms. There are many implementations of them to generate abstract images and music. Usually, the evolutionary program performs there the role of creator, confining a human to the role of critic.
On the other hand, I believe that humans naturally prefer to be creators rather than critics.
That was the motivation to devise a Human Based GA, that allows and encourages humans to take part in both roles. It is called human based genetic algorithm (HBGA) since all basic genetic operators are performed with the help of people. The algorithm processes strings of natural language and organizes knowledge flows within a community of individuals for the purpose of collaborative evolutionary problem solving, boosting innovation and creativity inside the community.
Social classifier structures. Optimal decision making in an organization.
classtre.gz(117k)
A simple model of organization as a classification cascade is proposed.
When a certain business plan or suggestion arrives at the organization, it
should classify whether the proposed project is useful or useless.
Every individual decision maker has a limited capacity to discern in advance
the potential value or harm, the proposed project would bring. Carefully
selected structure of decision making can reduce the risk of error even
for the case, in which individual decision makers are not adequate to reach
correct decisions. An approach to evaluation of decision making structures
is proposed. Nearly optimal structures suggested that surpass traditional
ones in classification quality.
Genetic synthesis of cascade structures
for particle classification
clasdise.gz(90k)
The goal of precise and reliable separation of a product into useful and useless parts
demands the cooperation of a set of separating elements. The grouping of them in
a cascade allows us to obtain an especially refined product.
At the beginning of this work there was an idea of synthesis of effective
classification cascades by the grouping of ineffective separation elements into
the structure.
When the number of the elements is large, characteristics of the cascade depend more
on its structure than on the quality of an individual separation element (classifier).
The search of the optimal structure becomes a difficult task because of a large number
of possible variants of interconnection between the elements.
The genetic method of synthesis allows us to determine a nearly optimal structure of a classifier
cascade. Particularly, a set of non-trivial structures was synthesized, with a better
quality of separation, compared with the cascades from the previous works on
particle classification.
A
Synthesis of static situative structures. Mathematical model
articl2e.gz(61k)
All activity of a man is devoted to constructing desirable relationships
between material and formal objects. In this work, an approach for solving a
narrower task is suggested: the synthesis of desirable relationships between
objects presented as black boxes with some inputs and outputs. The existence of
the relations is usually restricted by some conditions. These conditions defined
either locally (the rules of connectivity between inputs and outputs)
or globally (the criteria of existence of the whole structure, the set of objects
and relationships). The applications of this method range from tasks of barter exchange
optimizing (based on a set of commercial offers) to tasks of constructing
educational programs out of basic discipline modules.
Eliminating memory leak in Messy Genetic Algorithm (MGA-C) IlliGAL Report #98015
Messy Genetic Algorithm was developed in
Genetic Algorithms Laboratory of Illinois University. The algorithm has
many advantages compared to the simple genetic algorithm (SGA). Unfortunately,
the program (revision 1991) contained the errors in its C source, which made it too memory-consuming.
I found them in January 1997 because I could not run the program on my computer.
After the fix the program performed fine in the same environment. In 1998 David
Goldberg suggested me to write this technical report about the bugfix.
New articles will be available for download soon. Comments and suggestions
are welcome to
revision clasdise.ps.gz(95k) of
12/8/98 contains more advanced simmetrical cascade, the estimation of its
throughput and average number of classifications
The fixed sources can also be downloaded here.
Here is my favorite book on genetic algorithms!
.
