Self-Organized functional hierarchy through multiple timescales: Neuro-dynamical accounts for behavioral compositionality

Abstract

Based on the ideas of self-organized functional hierarchy in dynamics of distributed neural activities, we introduce a series of neural modeling studies. The models have been examined through robot experiments for the purpose of exploring novel phenomena appearing in the interaction between neural dynamics and physical actions, which could provide us new insights to understand nontrivial brain mechanisms. Those robot experiments successfully showed us how a set of behavior primitives can be learned with distributed neural activity and how functional hierarchy can be developed for manipulating these primitives in a compositional manner.