M.Sc Bartosz Lis
Institute of Computer Science
Technical University of Lodz

Library for Simmulation of Artificial Neural Networks

Contents

  1. Introduction
  2. Handling data
  3. Logging events
  4. Building a network
  5. Training a network

1. Introduction

1.1. Overview
This software is a library simmulating various neural networks initiated at Institute of Computer Science, Technical University of Lodz. Its main objective was to facilitate teaching neural networks in batches. During training neural networks can record their state (wieghts, inputs, outputs, training parameters, etc.) in log files. The log file format is both human and machine readable. Anything written to a log coud be from the same log restored (for example to feed another network).
1.2. Source location
Sources could be downloaded from ftp://ftp.ics.p.lodz.pl/pub/ai/ann.
1.3. Legal Issues
Usage and developement of this software is governed by the GNU GPL license version 2.0 or higher whichever You choose.
1.4. Authors
Copyright by
Name
 e-mail
 joined
 affiliation
Bartosz Lis
 bartoszl@ics.p.lodz.pl
 beginner of the project
 Institute of Computer Science, Technical University of Lodz
Jaroslaw Koszuk
 jkoszuk@ics.p.lodz.pl
 May 2003
 Institute of Computer Science, Technical University of Lodz
 

2. Handling Data 

Scalar values passed among neural elements are represented by the type double. Scalar values form aggregations. Arrays of weights and input/output vectors are implemented using class Term and TERM (defined in ann_term.h), which are concretizations for scalar double of template classes Array and ARRAY (defined in ann_array.h). Array (and Term) does not allocate memory for its scalars, which does ARRAY (and TERM). It must be ensured that objects of Term (or other concretizations of Array) have access to memory allocated by TERM object (or other ARRAY concretization).

3. Logging Events

4. Building a Network

5. Training a Network