mthree is a package for scalable quantum measurement error mitigation that need not explicitly form the assignment matrix, or its inverse, as is thus a matrix-free measurement mitigation (M3) routine.
M3 works in a reduced subspace defined by the noisy input bitstrings that are to be corrected. Because the number of unique bitstrings can be much smaller than the dimensionality of the full multi-qubit Hilbert space, the resulting linear system of equations is nominally much easier to solve.
It is often the case that this linear equation is trivial to solve using LU decomposition, using only modest computing resources. However, if the number of unique bistrings is large, and / or one has very tight memory constraints, then the problem can be solved in a matrix-free manner using a preconditioned interative linear solution method, e.g. the Generalized minimal residual (GMRES) or biconjugate gradient stabilized (BiCGSTAB) methods.