In the past week, I really enjoyed this paper. It’s about boosting, which arose as a solution to the question—“Can a set of weak learners create a strong learner?” The solution is pretty fascinating. You basically pit a bunch of weak learners against each other for a number of rounds, and pick the best one at the end of each round, updating the weights of the data points that the winner misclassified. You then combine all the weak learners as a linear combination into a strong learner, by means of their predictions.
I enjoyed it so much that I wrote out the algorithm in Python and really thought about his problems at the end of the paper. I was going to write a post about it, but I came to the conclusion that I can’t do a better job describing the topic than Raúl Rojas did.
Here is the link again, if you didn’t get it.