Login Register



Auto-login on future visits

Forgot your password?

Newsletter May 2012

How to use additional information to improve the classifier performance

Why meta-data matters?

Tutorial

What information is useful to design a classifier?
We all know that we need data and labels to train a classifier. But are labels enough? Usually we do have additional information, such as time of acquisition, frame name, material subclass, or patient ID. How can we benefit from this meta-data? Watch this 4 minute video to find out.

perClass course: Machine Learning for R&D specialists.

Announcements

training course on industrial machine learning

Do you want to design powerful classifiers for your machine?
Join us on 25-29 June 2012 at PR Sys Design office in Delft, The Netherlands.

This intense training course provides you with the practical methodology to develop your own solutions with perClass. Back to your office, you solve your challenge and demonstrate a working classifier to your colleagues!

Learn more register now

Impressions from March perClass course

Participants in the perClass training course for machine learning specialists

From the blog

The last perClass course took place on 26-30 March. The participants came from as far as South Africa. Some already had experience in pattern recognition and came to deepen their proficiency, others were new to the field.
Because of diverse application areas of the participants, we have discussed very different issues ranging from speed/performance trade-offs of real-time classifiers to handling categorical features in huge SQL databases. Some participants brought their own data sets which allowed them to connect the examples to their own problems. It has been a real pleasure for us to have such an active and interested group coming by!
Read some of the participants impressions

Custom sample visualization in interactive scatter plot in Matlab

Custom visualization in interactive scatter plot

Feature highlight

Would not be great if you could visualize the original data related to a misclassified sample?
You can do just that with the user callback option of sdscatter. Simply provide a Matlab function that will read out the original data such as image, video, or signal using the information available in your data set. The function gets called each time the user clicks in the interactive scatter plot. With the callback options you gain quickly understanding of your data.
See the example in the knowledge base