Autoencoders for Level Generation, Repair, and Identification

摘要: Autoencoders are neural networks for unsupervised learning and dimensionality reduction which have recently been used for generating and modeling images. In this paper we argue for the use of autoencoders in game content generation, recognition and repair, and describe proof-of-concept implementations of autoencoders for these tasks for Super Mario Bros levels. Concretely, we train autoencoders to reproduce levels from the original Super Mario Bros game, and then use these networks to discriminate generated levels from original levels, and to generate new levels as transformation from noise. We believe these methods will generalize to other types of two-dimensional game content.