NeurApp

NeurApp is an intuitive software for exploring approximation by artificial neural networks. It can be freely downloaded from the internet.

Contents:

About the Software
Free Download

System Requirements

Documentation
Screen shots
Other Software
Previous Versions

About the Software

The simple and intuitive software has been created by Igor Grešovnik and Tadej Kodelja. It is aimed at exploring properties and function of modelling by artificial neural networks. The software is based on the Investigative Generic Library (IGLib) and uses the Aforge.Net neural network library. See the User Manual and the Documentation Section for more details.

Documentation

NeurApp User Manual. PDF document containing instructions for installation, running, and using the software, and some background information about the software. Also available as html page and mhtml document.

NeurApp License Agreement. Contains terms of use of the software.

ReadMe file. Contains very basic information about the software.

NeurApp Code Documentation (base index). Code documentation for the software, automatically generated from documentation comments by Doxygen. Only the top-level part of code and the most relevant libraries are documented here. For additional documentation of base libraries, see the IGLib Code documentation.

See also:

IGLib home page. Contains information on the Investigative Generic Library, the code base for this software, including the complete code documentation (and wider code documentation).
My other documentation.
Documentation section on the page about my work at COBIK where I use IGLib as a base library for research software in the field of neural networks.
Docuentation section on the IOptLib page.
Collection of optimization-related links.
Ph.D. thesis on optimization shell Inverse.

Free Download

NeurApp can be downloaded here and can be used free of charge by anyone (see the license agreement). The current version is 1.0.

Download NeurApp (click on the link) or 32-bit version for older (32 bit) Windows systems.

Short installation instructions:

Download the compressed archive by clicking on the download link above.
Unpack contents of the archive to an arbitrary directory.
Run the executable from the location where files were unpacked.
That's it. For more information about how to use the application, run it, click on the "About" tab and then on the "Help" button, or read the user manual.

System Requirements

The application requires the 64 bit .NET framework or its alternative implementation such as Mono. One of these is available for most modern computing platforms and are many times pre-installed with the operating system. If not yet installed, you will have to download and install the .NET framework (for Widows operating systems) or Mono (cross platform) before using the software.
There may be problems with using the software on some older platforms, e.g. Windows XP. For more details, see the System Requirements Section at the IGLib pages.

Examples and Screenshots

Approximation of function of 1 variable.

sampled data for approximation of a function of 2
variables

Generating training data by sampling a user defined reference function of two variables.

sampled data for approximation of a function of 2
variables

Neural network - based model of a user defined function of two variables.Original function is shown for comparison, with errors shown in color scale.

detail - comparison of original function and ANN model

detail - comparison of original function and ANN model

Zoomed-in detail of an area with larger approximation errors.

Model with contours.

Plot of approximated data with contours shown.

Fine tuning display options.

There are a number of display options that make inspection of the approximation and its errors really easy and clear. One can individually switch on or off display of training points, grid, surfaces and contours of the original function and its approximation. Transparency of both surfaces can be continuously adjusted. In this figure, both original and approximated surface were drawn relatively opaque, therefore it is immediately clear where approximated function has higher or lower values than the original one. Contours of the original function are displayed, which gives better feeling of 3D shapes.

Other Software

NeurApp is based on Investigative Generic Library (IGLib).

Some publically available software based on this library includes:

HashForm - application for calculation and verification of checksums (MD5, SHA-1, SHA-256, SHA-512) for files and text.
HashShell - a shell application with various cryptography-related functionality (calculation of hash functions / checksums, symmetric and asymmetric encryption, certificate management, etc.).
NeurApp – an educational application for visually exploring features of function approximation with artificial neural networks (ANN). It creates 1D and 2D ANN models of user defined functions and provides visualization capabilities to compare these models with originals. It can be downloaded from Softpedia.
AnnApp – software for exploring multidimensional ANN-based models.
IGShell - a shell application based on IGLib.
Oscillator Modeller – an educational application for playing with nonlinear driven oscillators and their resonance curves.

Previous Versions

This software is available as freeware and is freely available on these pages. Different version can be downloaded below. For download and installation instructions, see the Download Section.

Version 1.3

Version 1.1

Version 1.0

Maintained by Igor Grešovnik

Updated in April 2015

Number of accesses since the last update:

If you don't see the access counter properly then please follow this link!