Neural Integrator is a visual
programming environment for prototyping and developing neural networks and other
directed graph based programming models.
Neural Integrator is not yet stable. When it is stable, a download will be available on sourceforge.
Neural Integrator (NI) is more than just another neural net toolkit. It
is a visual programming language and Integrated Development Environment
(IDE) with an extremely open architecture. You can easily connect Neural
Integrator projects to any input sources and output destinations
and displays. For example: Excel COM support, JPEG or other image formats,
and graphic displays are extensions, which, had we not included
them, you could have written yourself! You can also design your own
processing cells, or even wrap functions provided by other neural net
Neural Integrator is fast. The run-time engine and cell implementations
are written in C++. All network data types are checked at "compile-time".
Hence, python support does not decrease runtime performance.
Neural Integrator is flexible. You can design your own cell types in
C++, python, or as sub-maps. You can customize the graphical user
interface for your own cell types. Every aspect of Neural Integrator's GUI
and abstraction functions is open-source and customizable.
Neural Integrator is intuitive to use. With interactive drag-and-drop
cell controls, non-programmers can design sophisticated applications and
leverage the work of programmers.
Neural Integrator is all about programming flexibility and integration.
You can write custom cells that make calls to external libraries. You can
use multi-language programming to control or embed working networks, even
incorporate the networks into libraries. You can create (or contract for)
a custom network compiler that will encode your networks for your
In Neural Integrator creating support for Backpropagation Networks,
Elman Networks, Jordan Networks, Functional Link Networks,
Counterpropagation Networks, Multidirectional Associative Memories,
Hopfield Memories, Markov Chains, Fuzzy Neural Systems, Recurrent
Networks, and any variation or combination of these is a matter of
describing the appropriate functionality as properties of a map and/or its
component cells. In nearly every other neural net programming tool,
implementing such variety would involve re-coding the tool itself. Any
application that can be expressed with a discrete-time connectionist model
is within Neural Integrator's scope.
Once written, such independently created modules can be made available
for re-use through the Neural Integrator Open Source Network.
Neural Network Links
Networks in Hardware - Excellent overview. Who is using neural nets
and why, hardware vs. software, etc.
Neural Networks at
PPNL - Excellent site. Everything from Java demos to commercial
applications, and lots of good links.
The C kernel and Python interface are complete.
The GUI (wxPython) is complete but needs some detail work.
Module file saving and loading works (XML).
The standard library needs lots of work.
This button bar allows you to tour the WebRing.
We invite you to visit the other sites in the Python And Zope Community