 |
 |
User Defined Neuro-Dynamics
|
 |
 |
User-Defined Neuro-Dynamics (UDND) is an advanced research extension to
NeuralWorks Professional II/PLUS that supports the design of new neural network
paradigms that can then be incorporated into Professional II/PLUS and used in
deployed applications. Key neural network components that can be customized
with UDND include network topology, processing element neuro-dynamics, the
network control strategy, and learning/recall parameter definitions.
After the basic network architecture is defined, specific summation, transfer, learning
rule, noise, output, and error functions can be written in C and then compiled for use
in the architecture. UDND also supports checkpoint processing and special event
processing, so that links to external programs can be activated during either learning
or recall. All important network events, such as Network Loaded, Network Saved,
Learn Start, Learn Complete, Recall Start, and Recall Complete generate software
interrupts that can be trapped and handled in application specific ways.
|
|
|
Commercial Pricing
|
Platform/OS |
Code |
Price |
With 1 Year TAP Subscription |
| Intel x86/Windows |
UDPW-S |
$995 |
$1,145 |
| Intel x86/Linux |
UDPL-S |
$995 |
$1,145 |
| Sun Sparc/Solaris - 32 bit |
UDU2-S |
$1,995 |
$2,395 |
| Sun Sparc/Solaris - 64 bit |
UDU4-S |
$1,995 |
$2,395 |
| Silicon Graphics/Irix - 32 bit |
UDU2-S |
$1,995 |
$2,395 |
| Silicon Graphics/Irix - 64 bit |
UDU4-S |
$1,995 |
$2,395 |
| IBM RS6000/AIX |
UDU2-S |
$1,995 |
$2,395 |
User Defined Neuro-Dynamics Features
User Defined Neuro-Dynamics® (UDND) is an advanced optional extension of
Professional II/PLUS. UDND allows you to use the C programming language
to define your own learning rules, transfer functions, error functions,
and other aspects of processing in neural networks. These functions are
compiled and linked to produce Dynamic Link Libraries (DLLs), or shared
object files on Unix platforms, which are called by Professional
II/PLUS.
Fast Easy Design
Using UDND you can modify learning rules, have the new network up and
running in as little as 30 minutes, and know that it will work.
Creating a completely new network type takes longer, but you will
discover it is easier and quicker than other methods of network design.
Equally as important, creating a new learning rule or network this way
makes it possible to quickly and effectively compare it to other
supported network types. This is an essential advantage only available
with UDND.
Customize These Functions
In addition to the broad range of basic building blocks proved by
NeuralWare, UDND makes it possible to customize the following functions:
- summation
- transfer
- output
- error
- learn
- noise
- check point processing
- special event processing
When you have finished your customization, an single command compiles
and links your code into a DLL (or Unix shared object file) ready for
Professional II/PLUS. Properly constructed functions are also
compatible with NeuralWorks Designer Pack.
Only Write the Essentials
The UDND advantage is that it lets you modify only the building blocks
particular to your applications. In many instances it takes only a few
lines of code to construct a new paradigm. Through well defined and
strictly followed interface specifications, all of the other building
blocks fit seamlessly together into the new system. These greatly
reduces debugging time, because the only routines which need to be
tested are the new building blocks.
All of the capabilities of Professional II/PLUS, including network
editing tools, general graphing and analysis tools, and utilities are
available with the your new functionality.
|
|
|
 |
|