Title: Simulating and Reconstructing Neurodynamics with Epsilon-Automata Applied to Electroencephalography (EEG) Microstate Sequences

Abstract: We introduce new techniques to the analysis of neural spatiotemporal dynamics via applying $\epsilon$-machine reconstruction to electroencephalography (EEG) microstate sequences. Microstates are short duration quasi-stable states of the dynamically changing electrical field topographies recorded via an array of electrodes from the human scalp, and cluster into four canonical classes. The sequence of microstates observed under particular conditions can be considered an information source with unknown underlying structure. $\epsilon$-machines are discrete dynamical system automata with state-dependent probabilities on different future observations (in this case the next measured EEG microstate). They artificially reproduce underlying structure in an optimally predictive manner as generative models exhibiting dynamics emulating the behaviour of the source. Here we present experiments using both simulations and empirical data supporting the value of associating these discrete dynamical systems with mental states (e.g. mind-wandering, focused attention, etc.) and with clinical populations. The neurodynamics of mental states and clinical populations can then be further characterized by properties of these dynamical systems, including: i) statistical complexity (determined by the number of states of the corresponding $\epsilon$-automaton); ii) entropy rate; iii) characteristic sequence patterning (syntax, probabilistic grammars); iv) duration, persistence and stability of dynamical patterns; and v) algebraic measures such as Krohn-Rhodes complexity or holonomy length of the decompositions of these. The potential applications include the characterization of mental states in neurodynamic terms for mental health diagnostics, well-being interventions, human-machine interface, and others on both subject-specific and group/population-level.