Séminaire mOeX - Chris Reinke

Abstract

Many complex dynamical systems, artificial or natural, are still not fully understood yet. One such type of system are cellular automata, like the Game of Life or Lenia, which I will use as an example. They have been widely used as abstract models to study various aspects of self-organization and morphogenesis, such as the emergence of spatially localized patterns. However, findings of self-organized patterns in such models have so far relied on manual tuning of parameters and initial states, and on the human eye to identify interesting patterns. In this seminar, I give an introduction to intrinsically-motivated machine learning algorithms (IMGEPs) for the exploration of such systems. IMGEPs combine intrinsically-motivated goal exploration and unsupervised learning of goal space representations. Goal space representations describe the interesting features of patterns for which diverse variations should be discovered. I introduce an IMGEP algorithm that incrementally learns a goal representation using a deep variational auto-encoder, and the use of compositional pattern-producing networks (CPPN) for generating initialization parameters.

Relevant Paper: https://arxiv.org/abs/1908.06663

https://moex.inria.fr/seminars/2023reinke.html