This dissertation consists of a series of findings which all demonstrate the inability of basic visual models to generalize to alternate conditions. In particular, this dissertation studies the receptive field of visual neurons, a concept well documented in the literature.
The first two chapters are concerned with linear cascade models of retinal ganglion cell function in the primate. These models start with a linear description of the receptive field and have had demonstrated success with artificial stimuli such as white noise. How- ever, naturalistic stimuli represents a unique statistical regime and receptive field modeling schemes derived from artificial stimuli may not generalize. Utilizing population recordings of the macaque ganglion cells to a novel naturalistic stimulus, this dissertation seeks to clarify and quantify the generalizability. Chapter one demonstrates that the key linear receptive field stage is an unsuitable approximation of the ganglion cell response when probed by naturalistic stimuli. Chapter two demonstrates that simple physiologically based modifications of the receptive field models are also inadequate. These two chapters both demonstrate and quantify the limits of generalizing the models to natural scenes.
Chapter three switches to human visual psychophysics but also addresses the ques- tion of generalizability. The chapter examines the Westheimer Effect, a well documented visual detection effect thought to be mediated by the classic center-surround receptive field of visual neurons. The effect has been primarily documented with a single polarity of visual stimulation, and the author seeks to test whether the underlying center-surround receptive field model is able to generalize to alternative polarities of visual stimulation. Again, the finding is that model generalizability is limited and alternative mechanisms beyond the original model are present.
Chapter four returns to the primate retina and explores an effect not generated by the classic receptive field. It is a peripheral effect which the authors relate to a novel amacrine interneuron. A basic receptive field model is then augmented with an amacrine mediated peripheral term to match the data. The basic receptive field model would have been insufficient to explain the observations.