Update rereferencing.md

Author: arnodelorme

tutorials/05_Preprocess/rereferencing.md

@@ -47,6 +47,8 @@ undoing an average reference transform -- as long as you chose to
 include the initial reference channel in the data when you transformed
 to average reference).
 
+The Huber mean is a robust estimator of central tendency that reduces the impact of outliers by iteratively optimizing Huber’s loss function. This function acts quadratically for small errors, much like the arithmetic mean, but switches to a linear treatment for larger deviations, effectively down-weighting extreme values. The threshold, measured in microvolts, specifies the deviation from the mean at which channel values are treated linearly rather than quadratically. This approach is similar to the reference technique used in the PREP pipeline, although PREP additionally removes channels with excessive deviations—a step that isn’t necessary in EEGLAB if bad channels have already been eliminated using clean_rawdata. It is important to note that the Huber mean should not be used before ICA because it violates the linearity assumption; it is best applied after ICA.
+
 Note that the re-referencing function also re-references the stored
 ICA weights and scalp maps, if they exist.