Skip to main content

Everything we see is the result of light-induced patterns of electrical activity in the nerve networks of the retina and brain. These patterns are then processed into an intelligible form by complex transformations within these networks. Knight works to develop a description of these networks in mathematical equations that will allow scientists to predict how the system will respond to specific visual patterns.

Knight and his colleagues study the visual part of the central nervous system, applying technology and theoretical means refined in their own laboratory. The visual sense provides unique opportunities for highly structured input, making it a suitable system for understanding how the nervous system processes input information.

Knight is specifically interested in the broad multicellular interactions among biophysical processes, which individually lie at the subcellular level. His laboratory’s efforts involve the conjoined development of experimental procedures that induce and record detailed neural responses and theoretical tools, including dynamical equations and computer simulation, which may then describe those neural responses quantitatively.

Vision occurs when a moving pattern of colored light arrives from the external world and induces dynamical patterns of electrical activity in a sequence of neural processing networks—starting with the retina, which is specialized brain tissue, and continuing into the brain. At each step, profound signal transformations occur that ultimately reduce the input to a form useful for action. Knight and his colleagues study that process with computer-generated stimuli, including modified natural-scene movies, designed with theoretical considerations that facilitate the interpretation of response features in terms of the responding system’s predictive dynamical laws. Their data include single-cell recordings from identified nerve cells.

A detailed analysis of responses of retinal ganglion cells to time-dependent naturalistic stimuli have recently revealed that these responses may be classified within a special subset of neurons that might be descriptively called “faithful copy neurons.” A population of such neurons produces an aggregate firing rate that far better mimics its neural input than would different choices of neuron design. The way a network of such neurons behaves depends critically on the nature of interconnections and only insensitively on variations in the dynamics of the individual neurons. The generality of this feature and its implications for network design are currently under study.

Recently, Knight’s associates have been able to record simultaneously from numerous cells in the lateral geniculate nucleus. The responses of these cells naturally classify them into several categories. By the creation of a new data analysis technique, it has proven possible to quantitatively measure the rate of information transfer through a group of such cells and also to measure the degree to which information transmitted by different cell groups is independent or redundant.