Browse Prior Art Database

Neural Networks Portable Network Definition File

IP.com Disclosure Number: IPCOM000103302D
Original Publication Date: 1990-Sep-01
Included in the Prior Art Database: 2005-Mar-17
Document File: 1 page(s) / 70K

Publishing Venue

IBM

Related People

Bigus, JP: AUTHOR

Abstract

While the underlying process of training a neural network involves numeric computations, an effective user interface requires symbolic data representations. Disclosed is a method for providing customized user interfaces on top of a standard neural network development environment.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Neural Networks Portable Network Definition File

      While the underlying process of training a neural network
involves numeric computations, an effective user interface requires
symbolic data representations.  Disclosed is a method for providing
customized user interfaces on top of a standard neural network
development environment.

      Neural networks are massively parallel computing models.  They
have many features that make them similar to trainable statistical
classifiers.  The "training" process usually consists of presenting
pairs of numeric vectors (single-dimensional arrays) to the network.
This data is read from a file with a pre-specified format so a
generic development environment can access them.

      However, from a usability standpoint, it is desirable to
display the network state symbolically or graphically. While a neural
network development environment could provide standard representation
formats, the unique requirements of different neural network
architectures may require customized display formats.  By providing
an appropriate Application Programming Interface, a user can provide
customized routines which read inputs from the training data
(database) files, perform data encoding and decoding, provide input
and output display screens, and write data to print files for logging
the training process.

      The standard neural network training environment will call
user-written interface routines when the name of a custom interface
program is...