Mike Preuss (Dortmund University)
Experimentation (sometimes also called empiricism) is an important factor in domains with strong real-world influences. Many constraints and interactions within algorithmic factors and other game components and a final objective (player satisfaction) that is difficult to quantify make it hard to assess suggested solutions. The non-determinism introduced by many CI-based methods can be an asset but adds another source of uncertainty that requires a statistical approach. However, this noise factor cannot be completely get rid off here as long as there are humans actually playing the resulting games. We summarize the current state of research concerning this human factor and the currently employed methods to handle it. Furthermore, we attempt to take over some lessons recently learned in general CI concerning structured experimentation, and highlight the many possible pitfalls that can ruin an investigation if not handled properly. Research questions, design of experiments, and tuning are topics that turn out to be important in most experimental investigations. Finally, we touch on the currently underestimated use of (numerical) modeling and visualization in CI-affected games research. Statistics has recently made good progress in this area and the resulting techniques can only improve our current experimental methodology.
Mike is Research Associate at the Computer Science Department, University of Dortmund, Germany (since 2000), where he also received his Diploma degree in 1998. His research interests focus on the field of evolutionary algorithms for real-valued problems, namely on multimodal and multiobjective niching and the experimental methodology for (non-deterministic) optimization algorithms. He is also co-editor of the recent Springer book “Experimental Methods for the Analysis of Optimization Algorithms”, and co-chair of the
EvoGames event since 2009. He is currently working on the adaptability and applicability of computational intelligence techniques for various engineering domains, computer games and music. In games, he is most
interested in AI for realtime strategy games.
