When should I use REM?

Advantages

Readout error mitigation is a technique that deals with errors that many other techniques do not handle. For that reason it can often be combined with existing workflows to produce better results without much modification. The technique can also accept as much information as the user has about the measurement statistics in order to generate more accurate predictions.

Disadvantages

Readout-error mitigation requires the preliminary characterization of the measurement errors associated to a specific backend. Measurement errors can be expressed in the form of a confusion matrix and its estimation involves numerous state preparations and measurements in the computational basis.

The confusion matrix must then be inverted and provided as an input to the technique. The complete characterization of a full confusion matrix scales exponentially in the number of qubits. For a better scaling when generating these matrices, readout-error characterization can be simplified under the assumption that measurement errors are local with respect to individual qubits or group of qubits.