[question] How to use qiskit-machine-learning with real ibm backends?

[KonradKorus]

Hello,

I am using EstimatorQNN and NeuralNetworkClassifier for image recognition on quantum computers, following the tutorial here. I am trying to find a way to run this setup on a real IBM backend (not locally on a simulator). However, after spending several hours, I haven't found a clear solution.

My question is: Is there a way to use this approach on a real backend? If so, how can I pass training data to the backend, given that it runs in the cloud and not on my local machine?

From what I understand, I could transpile the circuit, but then I face issues with passing training data into the circuit (so I’m not even sure if it will run on the backend). Additionally, I have read that it might be possible to send multiple jobs to the backend, but this approach doesn’t seem to support backpropagation.

I hope you can help clarify this (or at least confirm if it’s not feasible so I can stop pursuing this approach 😄).

[edoaltamura]

Hi @KonradKorus, we do support hardware backends since v.0.8. The difference with the tutorial you're following is in the kind of Estimator primitive to pass in:

# we decompose the circuit for the QNN to avoid additional data copying
qnn = EstimatorQNN(
    circuit=circuit.decompose(),
    observables=observable,
    input_params=feature_map.parameters,
    weight_params=ansatz.parameters,
    estimator=estimator,  <--- This needs to be an qiskit-ibm-runtime primitive, and not StatevectorEstimator
)

At the top of the tutorial you'll need to replace

from qiskit.primitives import StatevectorEstimator as Estimator
estimator = Estimator()

with

from qiskit_ibm_runtime import QiskitRuntimeService, EstimatorV2 as Estimator
backend = <your real hardware backend>
estimator = Estimator(mode=backend)

You can find more info at the Qiskit IBM Runtime link here.

Note that if you want to run Estimator inside a Session, you'll need to embed the training loop in a Session environment yourself. Let us know if you need more help.

[edoaltamura]

You can also find more examples in the migration guide to help with real backends - you may replace GenericBackendV2 with your (real) backend from IBM Qiskit Runtime.