GECCO 2005 Workshop on:

Theory of Representations

Washington, June 25, 2005 (14:00PM-18:00PM). To be held as part of the

Genetic and Evolutionary Computation Conference (GECCO-2005)
Washington, D.C. USA June 25-29, 2004 (Saturday - Wednesday)
http://www.isgec.org/gecco-2005/

organized by:
Marc Toussaint, University of Edinburgh, UK (mtoussai@inf.ed.ac.uk)
Alden H. Wright, University of Montana, USA (alden.wright@umontana.edu)
Edwin D. de Jong, Utrecht University (dejong@cs.uu.nl)

The choice of representation crucially determines the performance of a heuristic search process. We believe that there have been very interesting new ideas and approaches on the subject of learning representations recently. However, a unifying point of view is currently missing and the different approaches are widely scattered in the literature with too little cross-fertilization. In this workshop we would like to gather such work and, in a discussion between the contributors from the various lines of research, fuse the various approaches and formalisms into a common framework. This framework might clarify what the scope of a theory of representations should be, what existing algorithms may be considered as cases of representation adaptation, and how the existing literature on the topic can be integrated in a broader picture -- thereby also seeking contact with related areas in Computer Science and Machine Learning.

Program

Download as pdf

Venue: Renoir room of the L'Enfant Plaza Hotel, on the second level

14:00
Introduction

14:05   Keki M. Burjorjee, Jordan B. Pollack:
Theme Preservation and the Evolution of Representation

14:30   Zbigniew Skolicki:
Survivability of Migrants' Genes in Heterogeneous Island Models

14:55   Edwin D. de Jong, Richard A. Watson, Dirk Thierens:
A Generator for Hierarchical Problems

15:20
Coffee Break

15:35   Marc Toussaint:
Factorial Representations to Generate Arbitrary Search Distributions

16:00   Franz Rothlauf:
Some Remarks on the Design of High-quality Representations

16:25   Alberto Moraglio, Riccardo Poli:
Topological Crossover for the Permutation Representation

16:50   Cezary Z. Janikow:
Adaptable Representation in GP

17:15
Coffee Break

17:30
Open Discussion

---

Call for Papers

We invite paper submissions on all aspects related to representations. This includes

• Reviews or general discussions of own work, work from a certain line of research, or historical work (e.g. done in theoretical biology) with the aim to contribute to a more integrative picture of what a Theory of Representations should encompass.
• Original, also preliminary, theoretical ideas or experimental studies and propositions of new heuristics when they support or point towards theoretically grounded principles.
• Also giving a presentation only may be acceptable. Please contact the organizers.

Important Dates

April 14, 2005	Electronic Submission Deadline
April 20, 2005	Author Notifications Sent
April 27, 2005	Camera-Ready Copy Deadline
June 25, 2005	Workshop, 14:00PM-18:00PM

Topics

Specifically, theoretical topics covered by the workshop will include any results on

• how problem characteristics change under transformations of the search space (including non-injective ones)
• how the effect of search operators (mutation & recombination) changes under transformations of the search space (keywords: compatibility, neutrality)
• criteria for good representations and how analyzable problem characteristics can guide the construction of good representations or operators (keywords: Linkage learning, Walsh analysis)
• the relation (e.g., equivalence) of adapting operators and adapting representations (including the equivalent rewriting of one algorithm as another one on another representation)
• the relation (e.g., equivalence) between modeling search distributions (e.g., EDAs) and the choice of representation
• the relation to other topics in Computer Science and Machine Learning, like Independent Component Analysis, Minimum Description Length principles, etc.
• understanding the evolution of the genetic code in Biology (the meaning of gene expression processes, epistatis, pleiotropy, complex phenotypic variability, etc)

From the experimental and algorithmic side, the topics covered include work on

• constructing, learning, or evolving representations, in particular hierarchical or compact codes, generative encodings (grammars, L-systems)
• switching representations; e.g., based on analysis of previous evaluations, implicitly by representation island models, etc
• modeling the evolution of the genetic code in Biology
• automatically defined functions, self-organizing representations, ontogenesis, indirect/recursive encodings, compact and hierarchical codes, gene expression, biological codes, latent variable probabilistic models, evolution of representations.

Submissions

All submissions should be send by e-mail in PDF format to mtoussai@inf.ed.ac.uk. Any length of manuscript is welcome. Please keep to the GECCO formatting guidelines.