Determine whether SamplerQNN postprocessing caused the flat hardware loss curves in Q-AGNN

Determine whether the SamplerQNN postprocessing data-loss mechanism affected the hardware results reported for the quantum-enhanced graph neural network for intrusion detection by Chaudhary et al., which used four virtual qubits on the 156-qubit ibm_fez backend with Qiskit ML 0.9.0.

Background

The paper examines a second published workflow whose configuration could trigger the identified SamplerQNN data-loss mechanism. Chaudhary et al. reportedly used SamplerQNN for hardware validation on ibm_fez, a 156-qubit device, with four virtual qubits and Qiskit ML 0.9.0. Such a configuration can cause the range filter to discard many valid shots, depending on qubit layout and measurement-register details.

The reported hardware and noisy-simulation experiments both produced flat loss curves, but the dataset was very small and the paper's authors did not have access to the raw measurement data. Consequently, the present paper explicitly leaves unresolved whether postprocessing data loss affected those results.

References

As with Ref. [18], we cannot confirm the issue affected their results without access to the raw measurement data, however, the setup matches the conditions for potential data loss in postprocessing.

— Impact of Data Loss in Postprocessing on Training and Inference of Quantum Neural Networks  (2609.05060 - Panambalom et al., 4 Sep 2026) in Section V, discussion of Ref. [19]