Noisy Measurement

One of the most noisy operations in superconducting machines is measurement. It’s a bit ironic that we don’t actually have a built-in operation for it!

I think we should define a NoisyMeasurementGate that has a confusion matrix defined. Then it should provide efficient implementations for each of the simulators.

It might be reasonable to extract the notion of applying a confusion matrix to a classical result into a zero-target gate, so the noisy measurement can decompose into measure+confuse instead of having to be defined as one thing. It also likely makes sense to incorporated some additional optional arguments into cirq.measure for creating noisy measurements.

[tanujkhattar]

From cirq sync:

Let’s proceed with option-5 proposed above to modify MeasurementGate.

[95-martin-orion]

@tanujkhattar’s option (5) appeals the most to me. invert_mask is a subset of the requested behavior; we just need to generalize that behavior to include confusion matrices. A separate NoisyMeasurementGate feels unnecessary.

My primary concern in this process is how qsim will interact with the result. qsim classical-state handling is entirely separate from Cirq’s log_of_measurement_results, so any change to how Cirq handles classical data requires a nontrivial change to qsim as well. Generalizing invert_mask works well with this, as qsim already has a code path for handling invert_mask-type behavior.