US Patent 10272341 Procedural level generation for games

Embodiments provide techniques for generating game levels based on estimated quality score. For example, a set of training data can be retrieved that specifies attributes and quality scores for each of a plurality of game levels. One embodiment trains a machine learning model for the gaming application using the set of training data. A plurality of generated game levels are generated using a procedural level generation algorithm. Embodiments can determine, for each generated game level, a respective quality score, using the trained machine learning model, and upon determining that the quality score for the generated game level exceeds a threshold level of quality, can select the game level for transmission to a client of the gaming application.