Neural Networks are an attempt to model what we understand to be the low-level functionality of organic nervous systems in the digital realm. Organic nervous systems appear to operate as multiple levels of stimulus-response mechanisms interconnected such that one cell might derive its stimulus from the response of another or a group of others.

It appears that organic systems have two major advantages, their being comprised of many independant but parallel processed units and the ability to learn. We attempt to simulate this by providing structures which are able to recieve input, perform very simple processing and provide output. The former yields a massively parallel (if somewhat slower) system while the latter yields trial and error, the ability to learn from mistakes and hollistic/conceptual approaches to problems. This ability is facilitated by the introduction of feedback; that is the process by which the system is modulated by the the response of the environment to its actions.

In computational models, we often attempt to solve both of these problems through some form of Object Oriented Programming (OOP). OOP provides independant encapsulated processing units and the ability to store their own state. Input parameters are pumped through a simple calculation biased by internal weights thereby producing output. The independance approximates the parallel processing found in organic nervous systems (even if each instruction is still handled sequentially in traditional CPUs). Meanwhile, each unit can modify its internal structure when appropriate. This is controlled by the creation of a feeback loop or a process which evaluates the surrounding environment (net positive or negative effect) and instructs the system to adjust the internal weights accordingly.

I designed and built this entire site myself (I know I'm no artist) with a goal of usability. Please let me know if you have any problems with it. For this or anything, please write to me at: micah_at_mail.csh.rit.edu