I would argue that learning can be reduced to statistical inference.