Friedrich-Alexander-Universität UnivisFAU-Logo
Techn. Fakult�tWillkommen am Institut für InformatikFAU-Logo
Artificial Intelligence Research Group




Research projects

Open Positions




Demos & Downloads



Solutions (in German)

Former Members






Dept. of Comp.Sc. >  Artificial Intelligence Research Group  > Demos & Downloads

Artificial Intelligence Research Group

Die Arbeitsgruppe Digital Humanities wurde vom Department Informatik anlässlich der Versetzung von Prof. Dr.-ing. Günther Görz in den Ruhestand zum 1.10.2012 eingerichtet. In ihr sollen Forschungsprojekte im Schnittbereich von Informatik und Geisteswissenschaften durchgeführt werden, insbesondere solche, die bereits an der Professur für Künstliche Intelligenz initiiert worden waren.

Die Professur für Künstliche Intelligenz wird als Professur für Kognitive Systeme weitergeführt und soll im Jahr 2014 neu besetzt werden. Der frühere Lehrstuhl für Künstliche Intelligenz existiert nicht mehr als eigener Lehrstuhl, sondern wurde auf Beschluss des Departments mit dem Lehrstuhl für Theoretische Informatik fusioniert, den Prof. Dr. Lutz Schröder seit dem Sommersemester 2012 innehat. Um die Vorgeschichte zu dokumentieren, sind an dieser Stelle auch die alten Seiten der Professur KI abgelegt.


Here you can download a performant POS-Tagger (Erlangen-Tagger) developped at our AI-Chair. Achieving up to 97.3 per cent of accuracy, the JAVA-protoype combines the popular HMM approach with a rule-based postprocessor proposed by Eric Brill in 1992. Thus, the tagger's language model consists of two components: A set of HMM-Parameter and a set of transformation rules. This data is automatically computed by the tagger by supervised training. Currently, there are two installation packages available (please follow the installation instructions):

Unix/Linux Bundle

Windows XP Bundle

Further more, as a result of training the tagger on 38.000 newspaper sentences, parameter / rule files for German can be downloaded:

HMM-Parameter (German)

Transformation Rules (German)

In addition to that, we provide the complete source code for integration in your own linguistic projects:

Java Source Code (Eclipse Project)

A user manual, mainly explaining the different parameters of the tagger, can be found here.

IMPORTANT: The software can be used "AS IS", without warranties of any kind. We shall have no liability for injuries, damages or any kind of loss caused by its use.


Here you can download a tool for morphological analysis of german words developped at our AI-Chair. Initially written in LISP, we now have a Java-Implementation.


usage: /path/to/jar/fau_morphology.jar <german wordform>

Java Source Code (Eclipse Project)


General Information:

The JADEOWLCodec is an OWL DL framework in the shape of a third party add-on for JADE (Java Agent DEvelopment Framework). It was created by researchers and students that are associated with the chair for Artificial Intelligence (KI8) at the University of Erlangen-Nuremberg (Germany).

JADEOWLCodec is distributed under the GNU GPL: The JADEOWLCodec is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version (2, June 1991) of the License.

The main idea (and the reason why we call it a codec) is to enable agents to send and receive messages with OWL DL as a content language. See this article from ESSLLI06 WS FOCA for one overview of the way in which OWL DL documents are used as the message content in ontology based agent-agent communication.

Core Features:

Knowledge Base: Unlike other content message codecs for JADE the JADEOWLCodec does not provide an alternative serialization of the classes from the jade.content.* package. Instead, it consists of an open world knowledge base implementation complete with reasoner based inference and consistency checks.

Runtime TBox: While the regular JADE Ontology Support relies on a static set of ontology classes (usually created from an OWL DL TBox with the Protégé Bean Generator plugin), a JADEOWLCodec knowledge base is not limited to the definitions known at compilation time. This opens new possibilities to experiment with agents that are learning new term definitions at runtime.

Wrapper Generator: The high level of source code readability that is achieved when using JADE Ontology Support ontology classes can be maintained by using optional wrapper classes that can be created from the OWL DL TBox definition. These provide easy access to property values and fillers through typed accessor methods. Because the wrapper classes are not the actual knowledge representation but a interface layer, the advanced possibilities of the JADEOWLCodec are kept, including the use of concepts added at runtime, consistency checks and reasoner based inference.

ABox Updater: An optional ABox updater component enables agents to consistently revise the instance level knowledge of their knowledge with the contents of incoming messages. This updater even allows basic operation without any application domain specific programming. Domain specific preferences can easily be implemented on top of this updater to get more specific results.

Documentation: We are currently writing a HowTo, which can be found here. The apidoc can is located here.



Further Information via email: Bernhard Schiemann

Required Software:

To develop JADE agents speaking in OWL DL and run the examples you additionally need the following software.
  • JDK 1.5 :
    • The current distribution of the JADEOWLCodec has been tested against JRE 1.5.0-10.
  • RACER :
    • RACER: binary distribution of the Renamed ABox and Concept Expression Reasoner

Künstliche Intelligenz am Finanzmarkt

Mit Methoden den maschinellen Lernes lassen sich unter anderem auch Ausfallwahrscheinlichkeiten von Krediten im P2P-Kreditmarkt lernen. Analysen, Implementierungen und Hintergrundinformationen zu diesem Thema finden sich auf Finanzhai.

Biochemische Anwendungen der Künstlichen Intelligenz

Die Herstellung guter Edelbrände wie Kirschbrand, Quittenbrand, Zwetschgenbrand oder auch aus schwierigen Rohstoffen wie Schlehen, Vogelbeeren oder Waldbeeren erfordert viel Fingerspitzengefühl. Bildet man jedoch Messreihen charakterischer Größen während der Destillation, lassen sich Muster ausmachen, die mit Methoden der Künstlichen Intelligenz (inbesondere des maschinellen Lernens) während des Destillationsvorgangs analysiert werden können. Zusammen mit der TU Weihenstephan wird gerade ermittelt, wie KI-Algorithmen eingesetzt werden können, um Edelbrand höchster Qualität computergestützt zu erzeugen.

Vorlagen für Studien-/Diplomarbeiten am Lehrstuhl (odt, tex)

zip file

  Impressum Last modified: 2014/3/25