Sparse fitness evaluation for reducing user burden in interactive genetic algorithm

Interactive evolutionary computation is a technique that performs optimization based on human evaluation, and we have proposed an image retrieval method based on the emotion using interactive genetic algorithm. This approach allows to search images not only with explicitly expressed keyword but also abstract keyword such as 'cheerful impression image' and 'gloomy impression image'. It searches the goal with a small population size and generates fewer number of generations than that of conventional genetic algorithm to reduce user's burden. But this property may derive local minimum and sometimes more poor solution than random search method owing to relatively small size population. In order to solve this problem, we suggest an idea of sparse fitness evaluation method using clustering method and fitness allocation method. This aims to allow not only to keep the advantages of interactive GA but also to improve the performance by utilizing large population.