Essentially, experiment with qutrits classifiers.
While for qubits there is a clear solution, consisting of the best linear classifier, since it is universally supported for on-board classification, for qutrits that is not enough, and native support is non-trivial.
Offline vs on-board
For this reason, it could be useful to experiment also with offline classifiers. But also trying to target on-board classification, when possible.
Storage
We also need to support the classification parameters storage. As a first iteration they could go in calibration.json, where they can be sourced by all calibration protocols.
Ideally, we may want to move the calibrated parameters in parameters.json, to use them also for circuit execution. Since we do not have a circuit execution model beyond qubits, this part is definitely not urgent.
Reporting
Of course, for whatever classifier we use, we will need suitable plotting and parameters display in the usual table.
Essentially, experiment with qutrits classifiers.
While for qubits there is a clear solution, consisting of the best linear classifier, since it is universally supported for on-board classification, for qutrits that is not enough, and native support is non-trivial.
Offline vs on-board
For this reason, it could be useful to experiment also with offline classifiers. But also trying to target on-board classification, when possible.
Storage
We also need to support the classification parameters storage. As a first iteration they could go in
calibration.json, where they can be sourced by all calibration protocols.Ideally, we may want to move the calibrated parameters in
parameters.json, to use them also for circuit execution. Since we do not have a circuit execution model beyond qubits, this part is definitely not urgent.Reporting
Of course, for whatever classifier we use, we will need suitable plotting and parameters display in the usual table.