Neural networks and Extrapolation

[Steve Wolstenholme]

Neural networks do not extrapolate very well beyond the training range
but EasyNN-plus has three methods of extending the training range.

New training examples can be generated by adding values at random
points within or beyond the existing training range while the neural
network is learning.

Multiple neural networks can be used to test forecasted values beyond
the training range.

Dummy training examples can be added that do not contribute to
training but allow validating and querying values to be entered that
are beyond the training range.

Using all three methods, the latest version of EasyNN-plus can
extrapolate inputs and outputs beyond the training range.

--
Steve Wolstenholme Neural Planner Software Ltd

EasyNN-plus. The easy way to build neural networks.

http://www.easynn.com