Models and dynamical analyses of neural systems for the Eriksen decision task.

摘要

In this dissertation I focus on the Eriksen task, a two-alternative forced choice task in which subjects must correctly identify a central stimulus and discard flankers that may or may not be compatible with it. I carry out dynamical system analysis on both a neural network model and a Bayesian inference model for it. I also extend the Bayesian inference model to study and simulate sequential effects in the Eriksen task. The scientific motivation is to understand neural mechanisms that underlie decision making tasks, including how conflicting flanker stimuli can interfere with information processing of the central stimulus, and how subjects allocate attention during task performance.; First, I study and analyze a connectionist model for the Eriksen flanker task. When solutions remain within the central domain of the logistic activation function, I show that analytical solutions of a decoupled, linearized model that is modulated by a pre-determined attention signal can provide reasonable estimates of behavioral data. I also show that the dynamics of the two-unit decision layer can be decoupled and reduced to a drift-diffusion model (DDM) with a variable drift rate, that describes the accumulation of net evidence in favor of one or the other alternative. I compare my results with numerical simulations of the full nonlinear model and with empirical data, and show that my results have a better fit, and use fewer parameters than the original model.; Two Bayesian inference models have been developed recently to model the Eriksen task. Both models are nonlinear, coupled discrete-time dynamical systems. I analyze the dynamics of those models by considering simplified, approximate systems that are linear and decoupled. I also investigate the continuum limits of these simplified dynamical systems, demonstrating that Bayesian updating is closely related to a DDM.; In order to study and analyze sequential trial-to-trial effects during performance of the Eriksen task, I propose a simple extension to the Bayesian inference model. Simulations of my extension model agree with the findings of human subject experiment. Lastly, I also show that the experimental data provide evidence of prior updating during trials.