Open Source

Accelerating K-Means on the Graphics Processor via CUDA – Source Code

Contact: Mario Zechner, mzechner[at]know-center.at, Michael Granitzer, mgrani[at]know-center.at

Abstract:
In this paper an optimized k-means implementation on the graphics processing unit (GPU) is presented. NVIDIA’s Compute Unified Device Architecture (CUDA), available from the G80 GPU family onwards, is used as the programming environment. Emphasis is placed on optimizations directly targeted at this architecture to best exploit the computational capabilities available. Additionally drawbacks and limitations of previous related work, e.g. maximum instance, dimension and centroid count are adressed. The algorithm is realized in a hybrid manner, parallelizing distance calculations on the GPU while sequentially updating cluster centroids on the CPU based on the results from the GPU calculations. An empirical performance study on synthetic data is given, demonstrating a maximum 14x speed increase to a fully SIMD optimized CPU implementation.

Details on Source:

The source code is available under GPL3 so feel free to do with it whatever is permitted by the license. The code was targeted at Windows machines, compiled with Visual Studio 2008. Compiling it under Linux shouldn’t be a big problem either though. There’s only a few windows specific parts (mostly the timer code) that can be easily replaced. As stated in the paper there’s rounding errors in the GPU implementation due to how the GPU handles multiplication/additions which are merged if possible using a different rounding mode and so on. Other than that i can’t think of any pitfalls at the moment. The code is not very well commented but should be mostly self explanatory. If there are any questions, we will try to answer them as good as possible and as time permits.

If you use or build upon our work, please give us according credits by citing

Zechner, M., Granitzer, M., Accelerating K-Means on the Graphics Processor via CUDA Proceedings of the 2009 First International Conference on Intensive Applications and Services (INTENSIVE 2009), IEEE Computer Society, 2009

Thanks also to Jan-Michael Frahm who provided some updates to the code.

gpuk.update.Jan 6,30 MB

APOSDLE – New ways to work, learn and collaborate

Contact: Stefanie Lindstaedt, slind[at]know-center.at

Abstract:
After the successfull completion of the EU research project APOSDLE, the Open Source Version of the APOSDLE Software is available now.

Know-Center released the first reference data set from the EU research project APOSDLE to promote the comprehensibility and generalizability of research findings in the field of Technology Enhanced Learning (TEL). Now with the open source version of APOSDLE available, this contributes to the success and value of “Work Integrated Learning (WIL)”.

The APOSDLE Open Source Package contains the APOSDLE Platform, the APOSDLE Client and all tools necessary to install and run APOSDLE. To provide a fast overview of APOSDLE’s variety of features, the package contains two sample domains from the field of innovation management and statistical data. The APOSDLE Open Source Package is licensed under GPL v3. For downloading the package go to http://www.aposdle.tugraz.at/results/downloads/opensource_packages

- APOSDLE – new ways to work, learn and collaborate!

For more information please visit http://www.aposdle.org

Caliph & Emir

Abstract:
Java & MPEG-7 based tools for semantic annotation and retrieval of digital photos and images.

Caliph & Emir are Java & MPEG-7 based tools for semantic annotation and retrieval of digital photos and images supporting structured, unstructured and semantic, graph like annotation and content based, metadata based and semantic image retrieval.Caliph and Emir are further developed in an open source project and can be visited and downloaded on sourceforge.net: http://sourceforge.net/projects/caliph-emir/

For further explanations, screenshots and videos showing the usage of Caliph & Emir visit http://caliph-emir.sourceforge.net/.

MoKi – A Wiki for Modelling both Topics and Business Processes

Contact: Viktoria Pammer, vpammer[at]know-center.at

Abstract:

Knowledge-based systems need formal models in which knowledge are stored in a computer-intelligible way. MoKi (Modelling WiKi) is a wiki-based modeling tool in which topics and business processes can be modeled. While typically such modeling tools are geared towards experts, we specifically want to make MoKi a “modeling tool for everyone”.
MoKi can be tried online. A storyboard describes how MoKi, together with an additional modeling tool, TACT, can be used to configure the work-integrated learning system APOSDLE.
MoKi is being developed together with the Data & Knowledge Management group at FBK in Trento, Italy.

SNGSVM Source & Executables

Contact: Mario Zechner, mzechner[at]know-center.at

Abstract:
Here we provide the source code, data sets and compiled executable files for the paper “A Competitive Learning Approach to Instance Selection for Support Vector Machines”. You will need Java 1.6 to run and compile the code as well as > 1GB of RAM to execute the testing program for large data sets. Please read the README file included in the package.

If you use the source code for experiments and publish research results accordingly, please credit our work by citing:

Zechner, M., Granitzer, M. A Competitive Learning Approach to Instance Selection for Support Vector Machines, Proceedings of the 3rd International Conference on Knowledge Science, Engineering and Management, Vienna, 2009

sngsvm 9,74 MB

TeA – Highlighting assertional effects of ontology editing activities in OWL

Contact: Viktoria Pammer, vpammer[at]know-center.at

Abstract:

It is difficult for humans to maintain an overview of existing knowledge and data already in reasonably mid-sized ontologies. For knowledge bases in OWL, we have developed the notion of assertional effects. Assertional effects illustrate knowledge that is gained (or lost) when axioms are added to (or removed from) an existing knowledge base.
Read more about the theoretical background of assertional effects.
Try online in MoKi how assertional effects can support you when editing knowledge bases in OWL.
The background functionality of computing assertional effects, given two knowledge bases, is published as TeA package (in Java) under the GPLv3. The TeA package offers functionalities to compare two OWL knowledge bases, and to compute assertional effects. The most extensive documentation of the TeA package is found here. An easy start to using the TeA package is to use the TeaTypes class and its getCollectedAssertionalEffects() method.
Source-Code and Binaries (compiled using Java 6)
Javadoc documentation.

• Source-Code und Binaries (Java 6 compiled)
• Javadoc-Dokumentation.