Recent Advances in Large Margin Learning