Sparse and reweighed learning. (With binned data.) These algorithms are probably slightly worse but they're useful for inference when number of samples is very

Sparse and reweighed learning. (With binned data.) These algorithms are probably slightly worse but they're useful for inference when number of samples is very small. If you apply them to dense data, the results are actually not bad. Not that much worse, either, and much stronger at generalization.