By default, min() and max() operate along the first non-scalar dimension. If you have a 2D array, m x n, with m and n both not 1, then that means that min() or max() of the array would produce a 1 x n output -- it has operated along columns, producing one result for each column.
Now suppose you transpose the m x n array to become n x m, with the rows becoming columns and the columns becoming rows, and you min() or max that. You will get a 1 x m result -- one result for each of what were originally rows.
Thus, min(M') is one way of producing a minimum for each row (but it has a problem if the data is complex-valued.)
More clear and robust is to use the syntax min(M,,2) to process dimension #2 specifically, producing one result for each row.