Reinforcement learning for quantum circuit optimization

One of the central bottlenecks for quantum computers is noise originating from their inevitable interaction with the environment, which threatens to corrupt the result of the quantum computation. The detrimental effect of noise can be mitigated by quantum circuit optimization, i.e., to search for circuits with less operations and a shorter runtime. In the recent years, powerful approaches [1, 2] have been developed focused on optimizing the global circuit structure. However, these approaches do not consider and thus cannot optimize for hardware specifics of the quantum architecture, which is especially important for near-term applications. To address this point, we propose a novel approach to quantum circuit optimization based on reinforcement learning, a powerful machine-learning technique which has already helped, for example, to achieve super-human performance in the board game Go [3].

[1] Kissinger et al. (2020), PRA 102(2), 022406
[2] Zhang (2019), arXiv:1903.12456
[3] Silver et al. (2016), Nature 529(7587), 484-489