Training-Free Quantum Architecture Search Under Realistic Noise via Expressibility-Guided Evolution

Area of Science:

• Quantum Computing
• Quantum Information Science

Background:

• Designing noise-robust parameterized quantum circuits (PQCs) is crucial for the noisy intermediate-scale quantum (NISQ) era.
• Current quantum architecture search methods are computationally expensive due to training large SuperCircuits and noisy SubCircuit evaluations.

Purpose of the Study:

• To develop a training-free quantum architecture search framework for noise-robust PQCs.
• To establish information-theoretic expressibility measures as effective surrogates for performance-based estimators in noisy quantum environments.

Main Methods:

• Proposed a novel framework utilizing information-theoretic expressibility measures, specifically KL-divergence, for quantum architecture search.
• Introduced an expressibility-guided evolutionary search that bypasses the need for SuperCircuit training and noisy executions.
• Demonstrated device-agnostic applicability by evaluating expressibility independent of specific hardware noise models.

Main Results:

• Empirically demonstrated a monotonic association between noise-free KL-divergence-based expressibility and noisy task loss across various architectures and noise models.
• Achieved competitive performance compared to SuperCircuit-based methods with substantially reduced computational cost.
• Validated the approach using IBM-derived Qiskit noise models.

Conclusions:

• Information-theoretic expressibility serves as an effective and computationally efficient surrogate for ranking PQC architectures under realistic noise.
• The proposed training-free, device-agnostic framework accelerates the design of noise-robust PQCs for NISQ devices.
• This approach enables reusable architectures across different quantum devices without re-searching.