DreamQAS: Model-Based RL for Quantum Architecture Search
A new framework called DreamQAS has been developed by researchers, utilizing model-based reinforcement learning for quantum architecture search (QAS), enhancing the efficiency of variational quantum eigensolvers (VQE). This approach, outlined in an arXiv paper (2607.29491), tackles the shortcomings of conventional RL-QAS methods, which inefficiently optimize VQE after each circuit extension, ignoring the deterministic aspects of circuit design. DreamQAS maintains these circuit dynamics and focuses solely on the costly post-VQE feedback. It employs a recurrent randomized-prior ensemble to estimate an oracle-free score against an empirical energy frontier, facilitating multi-step imagined policy learning across valid circuits. With a standard budget of 15,000 episodes, DreamQAS recorded the lowest mean frozen-policy energy error in four out of five molecular tasks, showcasing its potential to enhance quantum circuit design and advance quantum chemistry and materials science.
Key facts
- DreamQAS is a model-based RL framework for quantum architecture search.
- It preserves exact circuit dynamics and learns only expensive post-VQE feedback.
- A recurrent randomized-prior ensemble predicts an oracle-free score relative to an empirical energy frontier.
- It supports multi-step imagined policy learning over explicit legal circuits.
- Ranking-based activation, uncertainty-aware pessimism and truncation, and selective real-VQE verification form a reliability-controlled learning loop.
- Under a 15,000-episode budget, DreamQAS achieved the lowest mean frozen-policy energy error on four of five molecular tasks and second-lowest on one.
- The paper is available on arXiv with ID 2607.29491.
- The method aims to improve VQE efficiency in quantum computing.
Entities
Institutions
- arXiv