QuantumSEA: In-Time Sparse Exploration for Noise Adaptive Quantum Circuits

TL;DR

The paper tackles the challenge of training and deploying parameterized quantum circuits on noisy, time-constrained NISQ devices by introducing QuantumSEA, an in-time sparse exploration framework. It jointly optimizes a sparse circuit topology and gate weights under real device noise, leveraging a prune-and-grow strategy with a historical-gradient accumulator to guide growth. The method yields implicit circuit capacity, enhanced expressiveness, and superior robustness across QML and VQE benchmarks on multiple quantum devices, achieving notable reductions in gate count and training/execution time. This approach offers a practical path toward scalable, noise-resilient PQCs for near-term quantum computing with real hardware constraints, and demonstrates compatibility with existing noise mitigation techniques.

Abstract

Parameterized Quantum Circuits (PQC) have obtained increasing popularity thanks to their great potential for near-term Noisy Intermediate-Scale Quantum (NISQ) computers. Achieving quantum advantages usually requires a large number of qubits and quantum circuits with enough capacity. However, limited coherence time and massive quantum noises severely constrain the size of quantum circuits that can be executed reliably on real machines. To address these two pain points, we propose QuantumSEA, an in-time sparse exploration for noise-adaptive quantum circuits, aiming to achieve two key objectives: (1) implicit circuits capacity during training - by dynamically exploring the circuit's sparse connectivity and sticking a fixed small number of quantum gates throughout the training which satisfies the coherence time and enjoy light noises, enabling feasible executions on real quantum devices; (2) noise robustness - by jointly optimizing the topology and parameters of quantum circuits under real device noise models. In each update step of sparsity, we leverage the moving average of historical gradients to grow necessary gates and utilize salience-based pruning to eliminate insignificant gates. Extensive experiments are conducted with 7 Quantum Machine Learning (QML) and Variational Quantum Eigensolver (VQE) benchmarks on 6 simulated or real quantum computers, where QuantumSEA consistently surpasses noise-aware search, human-designed, and randomly generated quantum circuit baselines by a clear performance margin. For example, even in the most challenging on-chip training regime, our method establishes state-of-the-art results with only half the number of quantum gates and ~2x time saving of circuit executions. Codes are available at https://github.com/VITA-Group/QuantumSEA.

Figures (10)

• Figure 1: Overview of method comparisons. Both neural architecture search and pruning maintain dense circuits during training and produce their sparse variants in the end. And their training suffers from massive decoherence and operation noises due to excess quantum gates. Our proposal enables the circuit design from sparse to sparse, which only deals with a small portion of quantum gates at each training iteration.
• Figure 2: (Left and Middle) Real QC testing accuracy (%) over the number of parameters in quantum circuits. Evaluations are conducted on IBMQ-Santiago with RXYZ and MNIST-2. The accuracy of conventionally designed circuits is quickly saturated and then degraded. Our QuantumSEA outperforms other approaches in quantum noise mitigation, allowing larger circuit capacity and superior accuracy. (Right) Real QC testing accuracy (%) over the training time of quantum circuits, i.e., RXYZ) on IBMQ-Santiago.
• Figure 3: The overview of QuantumSEA. The upper figure demonstrates the in-time sparse exploration of quantum circuits. Specifically, it first trains sparse circuit for $\Delta\mathrm{T}$ iterations, then leverages prune-and-grow strategies to explore the crucial sparse topology, repeating until convergence. Both weight and topology updates are aware of real QC noises. The bottom figure presents an example sparse circuit for QML tasks, which consists of data encoder, trainable quantum layers, and the measurement layer.
• Figure 4: Example quantum circuits for VQE tasks.
• Figure 5: Real QC testing accuracy (%) over the number of parameters in quantum circuits. Evaluation are conducted on IBMQ-Santiago with configurations of {(IBMQ Basis, MNIST-2), (ZX+XX, Fashion-MNIST-2), (ZZ+RY, MNIST-4)} and real noise models wang2021quantumnas.