Machine learning: combinations of classification rules

We observe combined rectangles that approximate a curve. The boosting algorithm used to learn the ensemble of rectangles is either AdaBoost or GloBoost.

For these boosting methods, we use the Volata machine learning implementation.

100 boosting steps, 2 images per second: AdaBoost vs GloBoost

GloBoost with 100 boosting steps — GloBoost

AdaBoostwith 100 boosting steps — AdaBoost-MG

After 1000 boosting steps: AdaBoost vs GloBoost

GloBoost 1000 tours de boosting — GloBoost

AdaBoost 1000 tours de boosting — AdaBoost-MG

Noisy data, 100 boosting steps, 2 images per second: AdaBoost vs GloBoost

GloBoost noisy data and 100 boosting steps — GloBoost

AdaBoost noisy data 100 boosting steps — AdaBoost-MG