Modeling and optoelectronic realization of an artificial cortex
Abstract
Cortex, the outermost layer of the cerebrum, is recognized as the most developed part of the brain. It is believed that the higher-level functionality of the brain, the operations such as perception, cognition, and learning of both static and dynamic sensory information, originates from the dynamics of the massively interconnected gray cells of cortex. Because of the compact three-dimensional architecture of this biological computational paradigm, realization of bio-inspired machines that imitate such functionalities, including all the cellular details, is prohibitively difficult even if we consider the available nano-fabrication technologies. Based on this logical deduction, instead of considering each single neuron, an intriguing conjecture is to build aggregate level models that mimic the behavior of a population of neurons with collective emergent properties. In our approach, which is presented in this dissertation, cortex is assumed to be a composition of a sequence of discernable interconnected cortical patches. Each concerned patch is a network of asymmetrically coupled complex processing elements whose dynamics contain not only fixed-point and periodic attractors but also bifurcation and chaos. Dynamics of the complex processing elements, in this dissertation, is mathematically modeled by a slight modification of the time evolution of netlets adapted from computational neuroscience. Regarding this modification, the dynamics of a netlet is approximated by that of a quadratic return map. Studying the previous experimental observations demonstrates that a smart way of coupling such processing units is to couple them through their bifurcation parameters. Putting all pieces of this puzzle together, we model each cortical patch by a network of parametrically coupled quadratic return maps. Our simulations prove the ability of this network to emulate many salient features of cortical information processing, such as clustering, classification, generation of sparse codes and cortical topological computational maps. Next step in this research is seeking suitable enabling technologies, such as electronics and optics, for hardware implementation of these cortical models. It is a general consensus that realization of parallelism and massive interconnections can be done far better in optics compared to electronics. Nevertheless, one can exploit optoelectronic methodologies that combine the benefits of optics with flexibilities of electronics. An innovative optoelectronic approach is taking advantage of the optical mechanism of a special type of stimulable storage phosphor, the so called electron trapping materials. Our analytical modelings and experimental works reveal that the equilibrium state luminescence of this material can be controlled to generate a variety of different nonlinear behaviors including quasi-quadratic responses that can be used for generation of quadratic return maps. Combining this versatility with the state-of-the-art high speed spatial light modulators and CCD cameras, large arrays of quadratic return maps can be accommodated in a thin film of electron trapping material. Another approach which is investigated in this dissertation is based on using the recently developed digital microelectromechanic spatial light modulators. These modulators can control the exposure precisely. We show that a closed loop of such a spatial light modulator and a CCD camera can be used to build an optoelectronic machine suitable for parallel recursive computations similar to our cortical models.
Recommended Citation
Ramin Pashaie,
"Modeling and optoelectronic realization of an artificial cortex"
(January 1, 2007).
Dissertations available from ProQuest.
Paper AAI3292062.
http://repository.upenn.edu/dissertations/AAI3292062
View
More
