EvoWorkshops2006: EvoInteraction

1st European Workshop on Interactive Evolution and Humanized Computational Intelligence.

The founding works in interactive evolution, oriented towards artistic applications, have now been extended to many efficient and impressive applications where quantities to be optimised are related to subjective rating (as visual or auditive interpretation). Characteristic recent works concern for example hearing aids fitting, smooth human-like control rules design for a robot arm, design of HTML style sheets, or user-profiled web research engines.

Interaction with humans raises several problems, mainly linked to what has been called the user bottleneck, i.e. the human fatigue. This makes interactive evolutionary algorithms and, more generally, interactive computational intelligence, different from other (automatic) techniques. The human-computer interaction adds strong contraints to a computational process. Depending on how and for what it is designed, one then needs to tackle with real-time, (or at least user-real-time), small samples (with possible premature convergence risks for EAs), learning process, strongly guided operators, etc ...

This workshop is intended to address the various aspects of interactive evolution, and more broadly of computational intelligence in interaction with human intelligence, including methodology, theorerical issues, and of course new applications.

Topics of Interest:

Organising Committee

Program Co-Chairs
Evelyne Lutton
Evelyne DOT Lutton AT inria DOT fr
INRIA, France
Hideyuki Takagi
takagi AT design DOT kyushu-u DOT ac DOT jp
Kyushu University, Japan
EvoWorkshops2006 Chair
Franz Rothlauf
rothlauf AT uni-mannheim DOT de
University of Mannheim, Germany
Local Chair
Anikó Ekárt
ekart AT sztaki DOT hu
Hungarian Academy of Sciences
Publicity Chair
Steven Gustafson
smg AT cs DOT nott DOT ac DOT uk
University of Nottingham, UK

Note: the e-mail addresses are masked for spam protection.

Programme Committee

Thomas Baeck --- Leiden University / Nutech Solutions, USA
Eric Bonabeau --- Icosystem, USA
Fabio Boschetti --- CSIRO, Australia
Praminda Caleb-Solly --- University of the West of England, UK
Pierre Collet --- Unversite du Littoral, Calais, France
Michael Herdy --- Virtuelles Prototyping
Fang-Cheng Hsu --- Aletheia University, R.O. China
Christian Jacob --- Dept of Comp Sc., University of Calgary
Daisuke Katagami --- Tokyu Institute of Technilogy, Japan
Penousal Machado --- University of Coimbra, Spain
Yoichiro Maeda --- University of Fukui, Japan
Hiroaki Nishino Oita --- University, Japan
Ian C. Parmee --- University of the West of England, UK
Yago Saez --- Universidad CARLOS III de Madrid, Spain
Marc Schoenauer --- INRIA, France
Daniel Thalmann --- EPFL, switzerland
Tatsuo Unemi --- Souka University, Japan
Leuo-hong Wang --- Aletheia University, R.O. China

Accepted Papers: titles and abstracts

On Interactive Evolution Strategies
Michael T. M. Emmerich, Ron Breukelaar, Thomas B¨ack

In this paper we discuss Evolution Strategies within the context of interactive optimization. Different modes of interaction will be classified and compared. A focus will be on the suitability of the approach in cases, where the selection of individuals is done by a human user based on subjective evaluation. We compare the convergence dynamics of different approaches and discuss typical patterns of user interactions observed in empirical studies. The discussion of empirical results will be based on a survey conducted via the world wide web. A color (pattern) redesign problems from literature will be adopted and extended. The simplicity of the chosen problems allowed us to let a larger number of people participate in our study. The amount of data collected makes it possible to add statistical support to our hypothesis about the performance and behavior of different Interactive Evolution Strategies and to figure out high-performing instantiations of the approach. The behavior of the user was also compared to a deterministic selection of the best individual by the computer. This allowed us to figure out how much the convergence speed is affected by noise and to estimate the potential for accelerating the algorithm by means of advanced user interaction schemes.

Evaluation of Sequential, Multi-objective, and Parallel Interactive Genetic Algorithms for Multi-objective Floor Plan Optimisation
Alexandra Melike Brintrup, Hideyuki Takagi, Jeremy Ramsden

We propose a sequential IGA, multi-objective IGA and parallel interactive genetic algorithm (IGA), and evaluate them with a multi-objective floor planning task through both simulation and real IGA users. Combining human evaluation with an optimization system for engineering design enables us to embed domain specific knowledge which is frequently hard to describe, subjective criteria and preferences in engineering design. We introduce IGA technique to extend previous approaches with sequential single objective GA and multi-objective GA. We also introduce parallel IGA newly. Experimental results show that (1) the multi-objective IGA and the parallel IGA clearly provide better results than the sequential IGA, and (2) the multi-objective IGA provides more diverse results and faster convergence for a floor planning task although the parallel IGA provides better fitness convergence.

Creating Chance by New Interactive Evolutionary Computation: Bipartite Graph Based Interactive Genetic Algorithm
Chao-Fu Hong, Hsiao-Fang Yang, Leuo-Hong Wang, Mu-Hua Lin,Po-Wen Yang, Geng-Sian Lin

In this paper, our model supplies designing environment that used the component network to identify the high score components and weak components which decrease the number of components to build a meaningful and easily analysis simple graph. Secondary analysis is the bipartite network as the method for formatting the structure or the structure knowledge. In this step the different clustersˇ¦ components could link each other, but the linkage could not connect the components on same cluster. Furthermore, some weak tiesˇ¦ components or weak links are emerged by Bipartite Graph based Interactive Genetic Algorithm (BiGIGA) to assemble the creative products for customers. Finally, we investigated two significantly different cases. Case one, the cus-tomer did not change his preference, and the Wilcoxon test was used to evaluate the difference between IGA and BiGIGA. The results indicated that our model could correctly and directly capture the customer wanted. Case two, after the Wilcoxon test, it evidenced the lateral transmitting using triad closure extent the conceptual network, which could increase the weight of weak relation and retrieved a good product for the customer. The lateral transmitting did not present its convergent power on evolutionary design, but the lateral transmitting has illustrated that it could quickly discover the customerˇ¦s favorite value and recombined the creative product.

Practically Applying Interactive Genetic Algorithms to Customers Designs on a Customizable C2C Framework: Entrusting Select Operations to IGA Users
Fang-Cheng HSU and Ming-Hsiang HUNG

We propose a customizable C2C (customer to customer) framework to fully utilize interactive genetic algorithms (IGA) and to discover the potential capabilities of IGAs in customer designs. Traditionally, IGA users assign fitness to each chromosome. No matter the rating or ranking of the assignments, the traditional methods were unnatural, especially when IGAs were applied to customersˇ¦ designs. In this study, we find that allowing IGA users to directly select chromosomes into the mating pool according to their hidden fitness function(s) is not only a natural way to implement the select operations of IGA, but is also more effective. We call the model where parts of select operations are manipulated by users, the SIGA model. Preventing fatigue, however, is the most important challenge in IGA. The OIGA (Over-sampling IGA) model has been extremely effective at decreasing user fatigue. To verify the performance of the proposed SIGA, we conduct a case study and use the OIGA model as a benchmark. The results of the case study show that the proposed SIGA model is significantly more effective than the IOGA model. \end{abstract}

An experimental comparative study for Interactive Evolutionary Computation problems
Yago Saez, Pedro Isasi, Javier Segovia, Asuncion Mochon

This paper presents an objective experimental comparative study between four algorithms: the Genetic Algorithm, the Fitness Prediction Genetic Algorithm, the Population Based Incremental Learning algorithm and the purposed method based on the Chromosome Appearance Probability Matrix. The comparative is done with a non subjective evaluation function. The main objective is to validate the efficiency of several methods in Interactive Evolutionary Computation environments. The most important constraint of working within those environments is the user interaction, which affects the results adding time restrictions for the experimentation stage and subjectivity to the validation. The experiments done in this paper replace user interaction with several approaches avoiding user limitations. So far, the results show the efficiency of the purposed algorithm in terms of quality of solutions and convergence speed, two known keys to decrease the user fatigue.

Interactive Evolutionary Computation Framework and the On-chance Operator for Product Design
Leuo-Hong Wang, Meng-Yuan Sung, Chao-Fu Hong

Traditionally, product design problem is usually solved by means of the conjoint analysis methods. However, the conjoint analysis methods suffer from evaluation fatigue. An interactive evolutionary computation (IEC) framework for product design has been thus proposed in this paper. The prediction module taking care of evaluation fatigue is the main part of this framework. In addition, since the evaluation function of product design is an additive utility function, designing operators which heavily utilizes the prediction results becomes possible. The on-chance operator is thus defined in this paper as well. The experimental results indicated the on-chance operator can speed up IEC and improve the quality of solution at the same time.