Noise-Aware Quantum Architecture Search Achieves Robust Circuits with Nsga-Ii Algorithm

Why It Matters

By delivering noise‑resilient circuit designs, NA‑QAS reduces the error budget required for near‑term quantum applications, accelerating the deployment of practical quantum machine‑learning solutions.

Key Takeaways

• Noise model embedded directly in circuit training.
• Hybrid Hamiltonian ε‑greedy reduces evaluation cost.
• Variable‑depth NSGA‑II finds Pareto‑optimal architectures.
• Parameter‑sharing supernet cuts computational overhead.
• Demonstrated superior accuracy on noisy classification tasks.

Pulse Analysis

Noise remains the principal obstacle to practical quantum computing, especially for variational quantum algorithms that operate on noisy intermediate‑scale quantum (NISQ) devices. Traditional architecture searches often ignore hardware imperfections, leading to circuits that degrade rapidly in real environments. The NA‑QAS framework flips this paradigm by integrating realistic noise channels—such as bit‑flip and decoherence—directly into the training loop of parameterized quantum circuits, ensuring that candidate designs are evaluated under conditions that mirror actual hardware performance.

The technical novelty of NA‑QAS lies in three synergistic components. First, a hybrid Hamiltonian ε‑greedy strategy prioritizes promising architectures while keeping evaluation costs low, preventing premature convergence on sub‑optimal solutions. Second, a parameter‑sharing supernet allows multiple classical linear layers to share a common set of quantum parameters, dramatically reducing the computational burden of training each candidate individually. Finally, an enhanced variable‑depth NSGA‑II algorithm conducts multi‑objective optimization across performance, gate count, and circuit depth, producing Pareto‑optimal trade‑offs that respect both algorithmic expressibility and hardware constraints.

Empirical results on binary classification and iris multi‑class tasks under simulated noisy conditions demonstrate that NA‑QAS consistently yields higher fidelity circuits with fewer qubits and gates than existing methods. This efficiency translates into lower error rates and longer viable runtimes on NISQ hardware, bringing quantum machine‑learning applications closer to real‑world deployment. As quantum processors scale and noise mitigation techniques evolve, frameworks like NA‑QAS will be essential for automating the design of resilient quantum software, positioning industry and academia to capitalize on the next wave of quantum advantage.