Ensemble Learning
A committee of mediocre learners, correctly assembled, beats a single good one. The two ways to assemble it are the whole topic: bagging builds its members in parallel on resampled data to cut variance, and boosting builds them in sequence, each one aimed at what the last one got wrong, to cut bias.
The pages
- Ensemble Foundations - why combining weak learners works at all.
- Bagging & Bootstrap Aggregation - resampling, parallel training, and voting.
- Boosting & AdaBoost - reweighting the training set, round by round.
- Cheat Sheet - the whole topic on one page.