Inner retina interactions mediate edge detection. A CNN analysis of neuronal function

This paper describes two new techniques that help us understand retinal function. First, a cellular neural net (CNN) has provided us with an invaluable tool in the form of an hypothesis generator, with which we have been able to formulate our notions of complex retinal function, and express these hypotheses in “patterns of activity” describing the behavior and interactions of populations of retinal neurons. Second, we have developed methods for actually measuring patterns of activity in the living retina. This gives us the capability to verify our hypotheses. We can test our hypotheses by comparing the measured patterns taken from the physiology with the modeled patterns generated by CNN. To the extent that the predicted patterns are verified through physiological experiment, our measurements and hypotheses of retinal interactions are vindicated. Through the use of this tool we will be able to predict many emergent functional properties mediated by complex retinal circuitry that, until now, have not been considered or have simply been inaccessible to physiologists. Through this work we hope to be able to further define the principles of biological image processing and implement these principles in CNN design