Functional Architectures and Hierarchies of Time Scales

Dynamical system theory offers approaches towards cognitive modeling and computation inspired by self-organization and pattern formation in open systems operating far from thermodynamical equilibrium. In this spirit we propose a functional architecture for the emergence of complex functions such as sequential motor behaviors. We model elementary functions as Structured Flows on Manifolds (SFM) that provide an unambiguous deterministic description of the functional dynamics, while still remaining compatible with the intrinsically low dimensionality of elementary behaviors. Pattern competition processes (operating on a hierarchy of time scales) provide the means to compose complex functions out of simpler constituent ones. Our underlying hypothesis is that complex functions can be decomposed in functional modes (simpler building blocks). Simulations of generating cursive handwriting provide proof of concept and suggest exciting avenues towards extending the current framework to other human functions including learning and language.