Software Lab - Knowledge Extraction from Image and Engineering Databases

Year: SS 2013
Lecturer: Univ.-Prof. Dr. rer. nat. T. Seidl
M.Sc. P. Driessen
Dr. rer. nat. S. Fries
Type: Software Lab
Language: EN, DE
Content:  

In this summer term's lab, we offer the chance to get to know KNIME, a state-of-the-art data mining tool. While providing clustering, classification and analysis capabilities similar to Weka, KNIME also has the convenient and end-user friendly means thus far only been reserved for paying customers of commercial solutions like RapidMiner. Besides demonstrating the general potential of such mining tools, the milestones of our lab are for you to...

 

  • learn about various mining tools incl. their strengths and weaknesses [~2 weeks]
  • become acquainted with KNIME's intuitive Java-based extension system [~2 weeks]
  • implement your own KNIME plugins to extract and analyze information from our databases (or your own photo collection if you want) [~9 weeks]
  • briefly present your results, experiences and open questions of the development phase in a final demo session [1-2 weeks]

 

Screenshot of a typical KNIME workspace showing an examplary 2-part workflow.

 

KNIME uses so-called "nodes" to simplify and automatize complex data mining processes. The following screenshot outlines a complete workflow beginning with a "File Reader"-node to load data from, e.g., a CSV-file before applying three different colored visualization nodes in the upper branch and performing a k-Means clustering in the lower branch. The clustering results are visualized as well as stored within a new CSV-file by using the conveniently named node "CSV Writer".

 


    Complete mining workflow modeled in KNIME.
The "node-work" within KNIME is drag and drop based in order to enable the user to concentrate on the mining goal with only the right parameter choice remaining as a problem to be solved. As mentioned above, the underlying implementation of each node is realized in Java.
 
During the course of this lab, you will develop several small nodes on your own, each node providing a unique range of functions. Your implementations will be integrated into KNIME's open-source node repository and will - of course - be available from within the workspace as seen in the next screenshot.

 

In addition to the original nodes and the extensions available via the standard KNIME Update Site the community contributions offer an even wider functionality covering different application areas, such as chemo- and bioinformatics, image processing, or information retrieval.

Those of you who already have visited one of our data mining lectures may even recognize some of the folders and nodes shown in the screenshot, e.g., clustering and classification techniques. A very flexible type of node is presented at the bottom of the list, the "Java Snippet". As the name suggests, the user can apply these nodes to quickly fill a gap between two connected nodes in case a specific function is yet missing from KNIME's inherent node list. Inserting such a snippet would be the quick & dirty equivalent to the plugin approach we will pursue during the lab.

 

The lab as well as the individual supervision of the working group(s) will be held in English or German according to the whishes of the participants.

 
Hope to see you soon in our lab Sergej & Philip


    Extract from the KNIME node repository.

 

Dates:  
Type Date Room
Kickoff Meeting
Wed, 10.04.2013, 14:00 - 16:00
, Seminar room
Bachelor
Wed, 24.04.2013, 13:00 - 15:00
, Seminar room
Bachelor
Wed, 03.07.2013, 13:00 - 15:00
, Seminar room
Master
Wed, 05.06.2013, 15:00 - 17:00
, Seminar room
Bachelor
Wed, 05.06.2013, 13:00 - 15:00
, Seminar room
Master
Wed, 15.05.2013, 15:00 - 17:00
, Seminar room
Bachelor
Wed, 15.05.2013, 13:00 - 15:00
, Seminar room
Master
Wed, 24.04.2013, 15:00 - 17:00
, Seminar room
Master
Wed, 03.07.2013, 15:00 - 17:00
, Seminar room
Final Presentations
Wed, 17.07.2013, 09:00 - 12:00
, Seminar room