Curriculum reinforcement learning for quantum architecture search under hardware errors

• DOI: 10.48550/arxiv.2402.03500
• Publication Date: 2024-02-05
• Authors: Yash J. Patel; Akash Kundu; Mateusz Ostaszewski; Xavier Bonet-Monroig; Vedran Dunjko; Onur Danaci

The key challenge in the noisy intermediate-scale quantum era is finding useful circuits compatible with current device limitations. Variational algorithms (VQAs) offer a potential solution by fixing circuit architecture and optimizing individual gate parameters an external loop. However, parameter optimization can become intractable, overall performance of algorithm depends heavily on initially chosen architecture. Several search (QAS) have been developed to design architectures automatically. In case alone, noise effects observed dramatically influence optimizer final outcomes, which line study. search, could be just as critical, are poorly understood. This work addresses this gap introducing curriculum-based reinforcement learning QAS (CRLQAS) designed tackle challenges realistic VQA deployment. incorporates (i) 3D encoding restrictions environment dynamics explore space possible efficiently, (ii) episode halting scheme steer agent find shorter circuits, (iii) novel variant simultaneous perturbation stochastic approximation for faster convergence. To facilitate studies, we optimized simulator our algorithm, significantly improving computational efficiency simulating employing Pauli-transfer matrix formalism Pauli-Liouville basis. Numerical experiments focusing chemistry tasks demonstrate that CRLQAS outperforms existing across several metrics both noiseless environments.