A Discrete Holographic Model of Neural Memory in the Context of Invariant Perception and Recognition

184 p.Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1987.We present a model of biological memory based on the mathematics of Fourier holography. The scheme employs the well-established neural processes of spatiotemporal summation and Hebbian plasticity to perform the necessary operations. It is shown to be a content-addressable, associative memory whose output is pseudoinvariant with respect to spatial translations of the input. The distributed nature of the encoding algorithm and its embodiment of parallel processing make it attractive as a model of the brain.The relation of the holographic model to the popular correlation matrix memory is developed. A possible mode of self-organization of the network is outlined. Incorporation of the scheme into the primate visual system is found to yield global invariance with respect to rotation and scaling of retinal images. Results of computer simulations, which illustrate the properties of the memory and allow estimation of its capacity, are presented. These results suggest that holographic memory would be most useful in a setting where processing speed is more important than information capacity. However, unlike previous holographic models, ours may be modified to radically enhance its capacity while preserving its useful properties. We argue that this modification, known as dimensional shift restriction, is biologically plausible, and show that the signal-to-noise ratio for the modified network can be made arbitrarily large by increasing the dimensionality of the network. Other means of reducing noise in the network are also discussed.The phenomenon of metamerism, which refers to fact that there exist distinct spectral distributions of light which are perceived as being identical in color, is discussed. A theoretical basis is provided for an earlier, phenomenological model of this phenomenon, and its consistency with biophysical processes is demonstrated.U of I OnlyRestricted to the U of I community idenfinitely during batch ingest of legacy ETD