 So, as we just saw, if we have occasional outliers that are very big because in the mean squared error that error grows quadratically, these small number of errors will have a very large influence on what we finally get. That's an alternative, mean-absurd error. Robust estimation is an important field for people who don't know that field yet. It is, how can we do estimates in domains where occasionally we have outliers? And the idea is that we want the effect of every single data points to be bounded. Look at the loss function. If you take the derivative of that, you will see how this is bounded. So the loss function that we have in this case is the sum of the absolute values of the reconstruction errors. So now, take that same data set that has a few outliers and check what the outliers do in this case.