Criticality-facilitated decision-making in neural circuits: fractional diffusion in ongoing state and predicting individual decision-making ability from task-free resting state EEG

Decision-making is a fundamental component of human cognition. In behavioral two- choice decision experiments, reaction times (RT) vary across trials with a left-skewed distribution. In line with the highly variable neural signals, decision-making in such tasks has been modeled as diffusion process where stimulus-derived perceptual evidence accumulates as a drift within noise (i.e., random walk) towards a threshold at which the decision is made, followed by a response. Diffusion models of decision-making have become an important framework for neurocognitive science. Recently, models of coupled neuron circuits have been proposed to explain evidence accumulation in two-choice tasks as a winner-take-all in competition between neuron subpopulations. In these models, neuronal activity is assumed to show uncorrelated fluctuations around the mean firing rate. However, recent evidence demonstrates that neural activity fluctuations are much more than simple white noise; instead, they can be characterized as stochastic oscillations and critical neural avalanches with long-range correlations in space and time, emerging from the seemingly irregular spiking of neurons. An important question is how such highly non-trivial neural fluctuations impact on decision-making processes.

This project proposes to address this issue by combining nonlinear dynamics modeling and experimental data analysis. Our recent studies have shown that excitation- inhibition (E-I) balanced neuronal networks with biologically realistic synaptic kinetics can simultaneously encompass the ubiquitous irregular spiking, stochastic oscillations and critical avalanches and our novel mean field theory revealing that the network is operating close to a Hopf bifurcation from fixed point to oscillation. We will use the new network model and the derived mean field model to study decision-making and to reveal how critical neural dynamics could facilitate decision-making. Based on the mean field model, we plan to further develop a low-dimensional behavior model, analogous to the traditional diffusion model, which most likely will explicitly involve non-Markovian components describing fractional long-range correlations of ongoing brain states. Predictions of our models about the relationship between the fractal properties of task- free ongoing brain states and the decision-making performance will be tested in data from large-scale EEG experiments, by predicting individual differences in decision- making derived from RT distributions by individual differences in the variability of baseline neural fluctuations during resting states as personal trait. The project is expected to greatly deepen the understanding of brain-behavior relationships using approaches from nonlinear dynamics and could build foundation for developing potential dynamics-based sensitive biomarkers for the purpose of diagnosis and monitoring brain disorders and diseases.