Quantum machine learning and simulation: Practical insights for IT leaders

How qGANs work in practice

The architecture pairs a standard neural network generator with a quantum discriminator. The generator produces synthetic images, while the discriminator (running on a quantum circuit) compares them to originals and refines the results. Key steps include:

• Encoding the image into quantum states

• Running quantum operations (such as Fourier transforms or controlled-SWAP gates)

• Measuring output and decoding results back into classical form

Figure 1. qGAN hybrid architecture, classical generator supported by quantum discriminator

DXC’s team built this on AWS Braket with hybrid jobs; however, the approach is portable to other platforms, such as IBM Qiskit or CUDA-Q-based simulators.

Early findings

The experiments confirmed that hybrid quantum-classical setups can work today, even with limited quantum hardware.

Figure 2. Effects of qubits on hybrid optimization

Three takeaways:

• Convergence: Loss functions improved significantly within 200 epochs, with more qubits (4, 8, 16) speeding up convergence.

• Performance trade-offs: Local simulators were fastest but capped at around 20 qubits. Real quantum backends, such as those provided by Braket (like Rigetti Ankaa-3 and IQM Garnet), had slower runtimes and limited uptime.

• Hybrid dependence: Training leaned heavily on GPUs, showing that classical horsepower still carries most of the load while quantum provides targeted boosts.

The tests also highlighted a significant reality. Encoding and decoding data into quantum states eats into expected gains. However, that doesn’t erase the value; it just means hardware maturity and smarter pipelines will matter just as much as algorithms.

Challenges to bear in mind

Quantum is not plug-and-play yet. CIOs and architects weighing their options should consider the following:

• Hardware access: Real QPUs are scarce, with limited uptime and varying reliability across providers.

• Integration overhead: Data encoding and hybrid job orchestration add friction.

• Skills gap: Successful projects demand teams that blend quantum physics, ML and domain knowledge.

The path is clear. Each jump in qubit count and stability expands what’s feasible, moving quantum closer to mainstream IT strategy.

Put this knowledge to work

DXC’s work in the ABQCC and beyond shows that QML and simulation are no longer just science projects. They’re early-stage tools with real enterprise potential.

The quantum journey may be a bumpy ride, but the rewards could be transformative.

For CIOs and IT strategists, now is the time to get familiar, identify relevant workloads and build multidisciplinary teams that can translate quantum’s promise into better business outcomes.