GECCO-2001 Workshop

Non-Routine Design with Evolutionary Systems

7 July 2001

*** Schedule *** (new)

Description
Workshop Topics
Submission Details
Organizers
Workshop Format
Participation
Program Committee

Description

The book “Genetic Algorithms in Search, Optimization, and Machine Learning” helped bring the genetic algorithms to the spotlight zone and drawn the attention of a larger research community. However, its side effect is that the algorithms are distracted to the optimisation camp. To consider genetic algorithms (or evolutionary algorithms in more generic terms) as optimisation tool is appropriate only when we are trying to solve a routine problem. This kind of problems has the characteristics that all the problem variables and constraints are known in advance, and they do not change throughout the process. Once they are defined, an evolutionary algorithm is then applied to search the best combination of possible values under this specific instance of problem. Hence, the algorithm only comes to the problem-solving scene after a search space is identified “manually”. In other words, most of the research effort related to evolutionary algorithms is to use the algorithms as optimisers, i.e. to find “solutions as we already know”.

On the other hand, there are researchers who start applying evolutionary algorithms to non-routine problem, e.g. music, drawing, architectural design. For these problems, a significant portion of time is spent in identifying the variables and constraints. Depending upon the stages and the nature of the design, a global optimal solution may either not exist or is not necessary. The tradeoffs made under various circumstances create niches in the search space, where the best solution in one niche is hard to compare with the best performing one in another niche. Evolutionary algorithms have been demonstrated to be helpful in the realm of creative design from some of these projects. In fact, the evolutionary approach is a very good candidate to be a generator of “solutions that can be”. This is particularly useful in the early stage of a design and its potential is not fully exploited.

There are two issues highlighted. Firstly, an evolutionary algorithm is mistakenly to be treated as an optimiser. The misconception is due the fact that the algorithm is applied at the very last stage of a problem solving process, leaving the hardest part of problem formulation (mistakenly assuming to be trivial) to the human designers. As a result, the human designer’s knowledge and intuition of the problem limit the solutions thus found. Secondly, evolutionary algorithms are found useful in the non-routine design. One of the strength of an evolutionary algorithm can serve as ideas generator. Its potentials to derive creative solutions needed to be fully utilised. Design is a process where the problem space interacts with the solution space. Intermediate solutions created in the process help refine the current problem definition and tweaking the constraints. In most of the time, creative solutions are those which violate existing constraints.

What we try to address in this workshop is to maximise the usefulness of evolutionary algorithms by bringing them into a problem solving process much earlier, and to investigate how the early introduction of the algorithms may bring creative solutions. Therefore, the purpose of this workshop is not to report completed work, but to provide a forum to discuss and to generate research ideas, particularly related to non-routine design. The goal of the workshop is to see how quality and creative solutions can be generated for a mundane design problem by exploring what more an evolutionary algorithm can offer. It will be particularly useful and interesting to see how strategies and/or principles can be crossbred from previous attempts in different design domains.

Workshop Topics

The workshop topics include (but not limited to) the following:

Submission Details

Presentations will be selected according to the submitted papers. Please follow the main conference guidelines with regard to layout and length. Accepted papers will be published in the GECCO Workshop Proceedings. Important deadlines follow:
 
Paper submission deadline: March 2, 2001
Notification of acceptance: April 16, 2001
Final manuscript: April 30, 2001

Electronic submission (Postscript, or PDF) is strongly encouraged.  The organisers strongly recommend electronic submission of papers. An electronic copy of the paper (in either Postscript of PDF format) should be mailed to josiah@cs.usyd.edu.au xor mary@arch.usyd.edu.au  with subject "PAPER SUBMISSION" no later than March 2, 2001. Alternatively, authors may send four hard copies to one of the organizers.

Organizers

Mary Lou Maher (mary@arch.usyd.edu.au)
Key Centre of Design Computing and Cognition
U. of Sydney
Sydney, NSW 2006
Australia

Workshop Format

It is a half-day workshop. Authors are encouraged to submit position papers rather than completed work. People who have their papers accepted by the workshop will be given a copy of all the papers before the conference/workshop.

Each author is assigned to read another author’s paper in advance, and to prepare conjecture questions. Each paper presentation is like a debate. Author of the paper (representing the positive side) first gives a 5-min presentation to summarise his/her idea. It is followed by a 5-min of conjecture by the person who is delegated to read the paper and be the contender (the negative side). The author gives a 7-min response to the challenge. The floor is given a 15-min public discussion after the author's defense. A 20-min public discussion on issues related to the application of evolutionary algorithms to non-routine design at the end of all the presentations.

Participation

All the GECCO participants, especially those who are interested in design and creativity.

Program Committee