|Title:||Uncertainty handling in surrogate assisted optimisation of games|
|Abstract:||In this thesis entitled Uncertainty handling in surrogate assisted optimisation of games, we started out with the goal to investigate the uncertainty in game optimisation problems, as well as to identify or develop suitable optimisation algorithms. In order to approach this problem systematically, we first created a benchmark consisting of suitable game optimisation functions (GBEA). The suitability of these functions was determined using a taxonomy that was created based on the results of a literature survey of automatic game evaluation approaches. In order to improve the interpretability of the results, we also implemented an experimental framework that adds several features aiding the analysis of the results, specifically for surrogate-assisted evolutionary algorithms. After describing potentially suitable algorithms, we proposed a promising algorithm (SAPEO), to be tested on the benchmark alongside state-of-the-art optimisation algorithms. SAPEO is utilising the observation that most evolutionary algorithms only need fitness evaluations for survival selections. However, if the individuals in a population can be distinguished reliably based on predicted values, the number of function evaluations can be reduced. After a theoretical analysis of the performance limits of SAPEO, which produced very promising insights, we conducted several sets of experiments in order to answer the three central hypotheses guiding this thesis. We find that SAPEO performs comparably to state-of-the-art surrogate-assisted algorithms, but all are frequently outperformed by stand-alone evolutionary algorithms. From a more detailed analysis of the behaviour of SAPEO, we identify a few pointers that could help to further improve the performance. Before running experiments on the developed benchmark, we first verify its suitability using a second set of experiments. We find that GBEA is practical and contains interesting and challenging functions. However, we also discover that, in order to produce interpretable result with the benchmark, a set of baseline results is required. Due to this issue, we are not able to produce meaningful results with the GBEA at the time of writing. However, after more experiments are conducted with the benchmark, we will be able to interpret our results in the future. The insights developed will most likely not only be able to provide an assessment of optimisation algorithms, but can also be used to gain a deeper understanding of the characteristics of game optimisation problems.|
|Subject Headings:||Surrogate-assisted optimisation|
|Subject Headings (RSWK):||Optimierung|
Entscheidung bei Unsicherheit
|Appears in Collections:||LS 11|
This item is protected by original copyright
Items in Eldorado are protected by copyright, with all rights reserved, unless otherwise indicated.