Stuttgart Neural Network Simulator

The SNNS Project

SNNS is a software simulator for neural networks originally developed at the Institute for Parallel and Distributed High Performance Systems IPVR (Institut für parallele und verteilte Höchstleistungsrechner) at the University of Stuttgart, Germany. The goal of the SNNS project is to create an efficient and flexible simulation environment for research and application of neural nets, which is able to run on many computing platforms.

The SNNS simulator consists of the following main components:

• Simulator Core
• Graphical User Interface (originally X11 only)

SNNS also had a third component, the network compiler NESSUS. The input for the compiler is a network description in the proprietary language NESSUS. As output, the compiler creates a network file, which can be loaded by the simulator core. NESSUS is no longer maintained as far as I know.

Many people have contributed applications and extensions to SNNS like a version for Windows as well as for Java.

In 1991, SNNS was awarded the "Deutscher Hochschulpreis" (German Academy Award) for best education software in the field of Computer Science. The recipients of the award were Dr. Andreas Zell, our tutor, my fellow students Niels Mache (Simulator Core), Thomas Korb (network description language NESSUS) and myself (GUI for X11).

SNNS is now maintained at the University of Tübingen, Germany. See the SNNS Home-Page for further details.

My Contributions

• X11 User Interface (up to Release 2.2, 1990-1991)
• PostScript Generation (1992)

X11 User Interface

After conclusion of my studies, I continued to work on SNNS up to release 2.2 (March 1991) and implemented:

• Colored 2D-Display of neural network with graphical editor (Yes! The first version was only monochrome.)
• User Manual in English

Noteworthy feature of the GUI is the concurrent use of mouse and keyboard in the graphical network editor to perform most manipulations, which allows for fast operation by trained users. The mouse is used to select and point, the keyboard is used to determine the manipulation.

PostScript Generation (1992)

At the beginning of my professional career, while working at Eastman Kodak in Rochester/NY (USA), I needed to become familiar with the Adobe Postscript language. At this opportunity I developed a Postscript header (i.e. a set of macros and routines) for SNNS, which renders a neural network in various different ways depending on a set of input parameters.

Günter Mamier added a little "print driver" dialog to the GUI to provide these parameters to the header code and the actual output routine for the network elements.