Reciprocal Interactions of Computational Modeling and Empirical Investigation

Abstract

Models in general, and computational neural models in particular, are useful to the extent they fulfill three aims, which roughly constitute a life cycle of a model. First, at birth, models must account for existing phenomena, and with mechanisms that are no more complicated than necessary. Second, at maturity, models must make strong, falsifiable predictions that can guide future experiments. Third, all models are by definition incomplete, simplified representations of the mechanisms in question, so they should provide a basis of inspiration to guide the next generation of model development, as new data challenge and force the field to move beyond the existing models. Thus the final part of the model life cycle is a dialectic of model properties and empirical challenge. In this phase, new experimental data test and refine the model, leading either to a revised model or perhaps the birth of a new model. In what follows, we provide an outline of how this life cycle has played out in a particular series of models of the dorsal anterior cingulate cortex (ACC).