Login Register

Auto-login on future visits

Forgot your password?

Newsletter September-October 2010

How to build a detector?

Detector for one class


A detector is a classifier that focuses only on one class of interest. Training a detector means training a statistical model on the class of interest and adjusting the decision threshold. There are two principally different ways of adjusting the threshold:

  • We consider only the class of interest, and we adjust the threshold indirectly by choosing the fraction of training objects to be rejected.
  • We make use of other classes present, and set the threshold with ROC analysis.
Read more and see the example code in our Knowledge Base entry

New Release: PRSD Studio 2.2.4


PRSD Studio 2.2.4 (5-Oct-2010) is now available bringing these major features:

See release notes for the complete feature list.

We value your feedback, your questions/comments are most welcome in our forum.

Understanding outliers with interactive scatter

Visualization of outliers.

Feature Highlight

In order to have an idea of what might work for our pattern recognition task, understanding of our data is important. The interactive visualization of sdscatter in PRSD Studio comes to our aid in building this understanding.

In this short video we show how this simple interactive tool helps us to realize what causes the outliers in one of the classes, and how we can easily remove them.

See the video in our blog

Advanced Pattern Recognition course


Would you like to deepen your knowledge of patter recognition? The Pattern Recognition Lab at TU Delft and us, at PR Sys Design, join the effort to offer you a five-day course that combines both theoretical and practical insight. The APR course will help you to extend your competence and practical capabilities. When? On 25-29 October 2010, at TU Delft university, The Netherlands. Here you find detailed info.

The course is almost fully booked... register now!

ICPR 2010 in Istanbul

ICPR pictures

This year the International Conference on Pattern Recognition (ICPR) was in Istanbul, Turkey. The conference location was very nice. We really enjoyed meeting old and new colleagues. With several parallel tracks, there was plenty of research ideas to learn and discuss. Many people were interested in our poster on optimization of classifier hierarchies. In fact, we had so many visitors to our poster that had no time to walk around the other posters of the session!

Of course we also enjoyed the warm atmosphere of Istanbul, with its historical buildings, lively bazaars and delicious food!