 |
|
| |
GNOSIS – Polynomial Neural Network
Modeling
Many scientific, engineering, marketing, and economic phenomena
are complex and difficult to model. Often, a large number of
conditions that influence the input variables or an incomplete
understanding of the underlying principles governing the system
exacerbate the problem. The Barron Associates’ Generalized
Networks for the Optimal Synthesis of Information Systems (GNOSIS)
software suite provides a versatile and powerful method for
modeling a variety of complex phenomena. GNOSIS can solve diversified
problems such as:
modeling systems with continuous-valued
outputs (e.g., what is the temperature inside a reactor?)
classifying data into two or more categories (e.g., will the
market group, down, or stay the same?)
predicting the future values of time-series data (e.g., what
will tomorrow's temperature be?)
GNOSIS offers a number of technical advantages over both traditional
statistical modeling and other neural network software tools.
These include:
building blocks that can, on
their own, model complex nonlinear datainter relationships
a network training algorithm that is orders of magnitude faster
than the gradient-descent methods employed by most neural
network algorithms, and
a 'C'-language code generator that allows the user to compile
and incorporate networks in custom applications on any computer
Additionally, with the optional Algorithm for Synthesis of Polynomial
Networks - III (ASPN-III) structure-learning feature, GNOSIS
can rapidly sort through thousands of candidate inputs and structures
to find a feedforward estimation network of just-sufficient
complexity that is designed to perform optimally on unseen data.
The structure-learning capability can significantly reduce the
time required to obtain a good model, while enhancing model
robustness.
|
|
