| |
Building Neural Pathways in Real-Time: Neural Regioning Technique
Last Update: Thursday, May 13 2004 05:30 AM
jonathan@nuclearelephant.com
Download the White Paper (LNCS)
When I first wrote this paper, I didn't realize I was basically mirroring
existing research on evolutionary algorithms. I did realize, after some
additional research, however, that the randomness of evolutionary
algorithms are their fallacy, just as is the case of the alleged theory
of evolution. The remaining innovation I provide in this paper stems from
my beliefs in an ordered creation, rather than evolution, of man. This
serves to submit a possible alternative to the initial stages of an
evolutionary algorithm by removing randomness and supporting ordered
construction based on some type of base sterotypes and characteristics of
the data. While randomness is mentioned, beginning with order is, I believe,
the key to perfecting evolutionary algorithmss.
Neural networking provides an extraordinarily powerful parallel processing architecture for the recognition of complex patterns, concepts, and associations. As we continue in our attempts to understand the complexities of the biological brain, computer scientists search for ways to implement artificial neural networks capable of building interconnections using finite processing capabilities. A common bottleneck in programming a neural network is the exponential number of interconnections stored tentatively to establish future neural pathways, which frequently results in the need for pre-training of historical data into a hash or other structure for correlation. Neural regioning is a theoretical approach to establishing neural pathways in real-time based on a psychological approach of how humans form similar types of interconnections in everyday life - using a region correlation technique, which progresses from large, vague concepts into smaller, more specific neuron correlations without generating much peripheral data outside of the region's focus. It is essentially a neural networking filter designed to establish a basis for analyzing neural interconnections. This technique allows the creation of neural pathways to be created without any historical training data, building incrementally in real-time and on many different tiers ranging from individual neurons in a network to associating large regions. It is directly contrary to many such approaches where an initial implementation of data is stored in a grand scale, later drilling down on key data and discarding peripheral data.
|
