Deep Learning Novelty Exploration (DeLeNoX) approaches evolutionary divergent search through the perspective of intrinsic motivation, considering what a generative algorithm has previously produced when calculating what constitutes novel content. This is achieved by assessing novelty in terms of a higher-level representation, determined through representation learning algorithms (such as autoencoders) trained on a dataset of past products of the same algorithm. By allowing the generator to adjust its measure for novelty according to its own observations, DeLeNoX is able to continually adapt its focus to search beyond its current biases, a concept which is critical for achieving open-ended evolution. DeLeNoX has been applied for the generation of 2D spaceships and 3D Minecraft buildings, and has inspired a plethora of work on evolutionary computation.

Relevant Publications

  • Antonios Liapis, Hector P. Martinez, Julian Togelius and Georgios N. Yannakakis: "Transforming Exploratory Creativity with DeLeNoX," in Proceedings of the Fourth International Conference on Computational Creativity, 2013, pp. 56-63. PDF BibTex

  • Matthew Barthet, Antonios Liapis and Georgios N. Yannakakis: "Open-Ended Evolution for Minecraft Building Generation," in IEEE Transactions on Games, 2022 (accepted). PDF BibTex