the next learner in classifying them correctly?
Perhaps because it increases the discrepancy of outliners in comparison with the trend?
but how? let's say it's a binary dataset, and one of them was 1 but was classified as 0, how does increasing weight on this particular data make the classifier predict it as 1?
Even though it's a binary dataset, the weights aren't binary. Increasing the weights should just increase the distance between the points, not change the trend itself. Hence the outliners will deviate farther from where they are expected to be if they were correctly classified.
Обсуждают сегодня