Neural System Identification

Abstract

One could argue that all scientific problems can be described in terms of two fundamental objectives: identification and control. Much of the current body of research in basic neuroscience revolves around the problem of identification, although not formally posed as such. The problem of identification is that of cause and effect. For example, in considering the relationship between two synaptically connected neurons, how does the presynaptic action potential cause the postsynaptic potential? At a more macroscopic level, how do the photons of light entering the eye cause the neuronal population activity in the visual pathway of the brain? Due to the overwhelming complexity of the nervous system, it is in fact difficult to think of threads of investigation that are not in some way reliant on identification or modeling at the systems level. The concept of system identification goes beyond simply reporting experimental observations. In many cases, the input can be controlled, and the goal is to identify a functional relationship between stimulus and response that will enable prediction of the response of the system to subsequent arbitrary inputs. Failure in prediction exposes previous misconceptions about the underlying dynamics, often leading to more intelligently designed experiments, and so on. Herein lies the true value of the identification process in this largely empirical field of science.