Login Register

Auto-login on future visits

Forgot your password?

Newsletter February 2013

Cross-validation use and advantages

Choose a classifier with cross-validation


How to reliably estimate performance of a classifier?

In this tutorial we illustrate how to reliably assess classifier performance using cross-validation. We explore evaluation strategies in a computer vision application with apriori known structure (multiple samples per video sequence).

Watch this 3 minute video to find out more.

perClass course: Machine Learning for R&D Specialists.


training course for industrial machine learning

Want to learn how to build the best classifier for your problem? Join our very international group of participants!
Few spaces are still available in the 11-15 March 2013 course in Delft, The Netherlands.
This intensive training course provides you with the practical methodology to develop your own solutions with perClass. Back in your office, you solve your challenge and demonstrate a working classifier to your colleagues! Read testimonials

Learn more register now

Delft Robotics Symposium

Robots and stand at the event

From the blog

On January 25th 2013 the kick-off event of the Delft Robotics Institute took place. All the robotics labs of Delft Technical University joined together showing their research results. There were robots that walk, dance, cook, play soccer; the elephant robot for playing with hospitalized kids...
We were also present, as one of the spin-off companies, demonstrating our technology. We showed two industrial quality control demos. However, general public was attracted mostly by our live gender classifier labeling faces detected by OpenCV as "male" or "female". It was fun to test the classifier on such a sensitive issue... We were quite satisfied with the classifier reliability, but you know, it was not always right :)
See our demo video, and the event coverage by Dutch national TV news or by Tweakers.

Better classifiers with regularization

Feature highlight

Regularization example on hand-written digits

How to build classifiers in a small sample size problem with larger number of features? Gaussian models suffer in this situations, but may be still constructed with regularization.
perClass provides an easy automatic way to regularize covariance matrices. With an example of handwritten digits we illustrate that classifiers may be still built in high-dimensional spaces when properly regularized.
Read more in our documentation on quadratic classifier.